<%BANNER%>

Miniaturization of Ground Station for Unmanned Air Vehicles

xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20110115_AAAACN INGEST_TIME 2011-01-15T17:00:28Z PACKAGE UFE0008480_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 437537 DFID F20110115_AABUBS ORIGIN DEPOSITOR PATH rodriguez_u_Page_05.jp2 GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
519a7b432b7210cd3cc78b028d3ac336
SHA-1
4af78abdfc589bfe2c18e623416ee11734436e7e
4632 F20110115_AABUGP rodriguez_u_Page_41thm.jpg
ec22b143be3a12bf72129110ecd9ee05
c668ab1f8154760e361394718a18d3ef83b6eb90
73116 F20110115_AABTZR rodriguez_u_Page_12.jp2
3628bc3a2d008cd4f1323d39358ca3c3
58d6ac7af860cc53b63301cb056d21de09972ca3
74733 F20110115_AABUBT rodriguez_u_Page_06.jp2
de395a917116e14f4f5f27b94736dcfd
c5522fe508b6bba12bd438ff52610f99f3b4dfb2
41961 F20110115_AABUGQ rodriguez_u_Page_32thm.jpg
d5b5ad752ccfcb789099b43e0bacdbed
667507235aef82769b745ab9c5b6345e7e2e13fd
27759 F20110115_AABTZS rodriguez_u_Page_38.pro
4d9fea6c9b034ae6882de000f3149539
a073f1f1112485fba129c35252bc439f6d4bbc48
84533 F20110115_AABUGR rodriguez_u_Page_19.QC.jpg
827b7c6cfa4e322323132e80475bfa71
7507c167042150b60fd783ec49a2530b06325722
59028 F20110115_AABTZT rodriguez_u_Page_26.QC.jpg
582e36127ef8356cc0c4803ed337e3fc
66c95bff1145438f3cdbeef3c6740cdb062567f5
43496 F20110115_AABUBU rodriguez_u_Page_07.jp2
693ef43cbf3d4b29be573cd5c91f909e
d8462607977d02e0cfcc92ceca9a44e642bfb710
42742 F20110115_AABUGS rodriguez_u_Page_20thm.jpg
06626f8a311f8d8885ec05eda6833587
971c01c358ee1821c25411e51d5932df03b9aedf
35075 F20110115_AABTZU rodriguez_u_Page_27.pro
e1bc885c66de0d7bd58151e0ef9f8cee
ebf3dc0cb9fead2a3ea3bdd78e89dc513296223e
82110 F20110115_AABUBV rodriguez_u_Page_08.jp2
3a941244cbf1767fd7acadb8cc635d1d
d6b1cf5d0194bfa74c59a9c5534a57a1da1c4c9c
16094 F20110115_AABUGT rodriguez_u_Page_11thm.jpg
867eaa8c3b3c4d4aac5fdf28bd847fe1
e0296c3785327cf4fb9db634fca02f5c7986aa40
1053954 F20110115_AABTZV rodriguez_u_Page_17.tif
1054fd2caf010a212467dcab7e77d28a
d17313896a37312dc332ea287e77573bbc7246af
67034 F20110115_AABUBW rodriguez_u_Page_16.jp2
cf10c501767838187b9b83c87e095ceb
266a54980a3b689cdd353e5d8638c511dad115e7
69319 F20110115_AABUGU rodriguez_u_Page_31.QC.jpg
99345e043531b74bb21b8cedc1ff3e5e
05383c146b8b1b4336c36eafcca5883fe2919d3c
56383 F20110115_AABTZW rodriguez_u_Page_27.QC.jpg
60a1a0de7cf863565f3ab1223f7dab1d
edb089e96b2bb290d2d8555f44ba73f2865efbc5
84879 F20110115_AABUBX rodriguez_u_Page_17.jp2
103fbc0bd10b09998f8e55032357ad30
efbf35f099818306a9f598795aacf61115276075
42246 F20110115_AABUGV rodriguez_u_Page_18thm.jpg
8325ac35dfda769bf6b7b51aa89c5f8b
023c91e2ebf833654579143550f7f1a029f06a68
693119 F20110115_AABTZX rodriguez_u_Page_32.jp2
25e23846864ace3ad1f2361a69f6b500
ebe93994d5bfa1218417b68f0a7bad16de090d29
20887 F20110115_AABUGW rodriguez_u_Page_14thm.jpg
64403843c64c964c9c1872082776d21e
2b9ef553ed0b29fab0eefbd301b989bcc930d9cb
1386 F20110115_AABTXA rodriguez_u_Page_11.txt
1d926e5839b870c2510651fa158579e5
b27ffe2543d86e16063947cf04a3448f0b5044a9
25271604 F20110115_AABTZY rodriguez_u_Page_38.tif
5410cccf9532ed31c6316be431181d6b
e0c46f1c65d7e43052d73575b4517f353acab2de
935860 F20110115_AABUBY rodriguez_u_Page_18.jp2
ccd4b46b4d0aea85e3f973e83b523c7a
7f8d4762ad6d41893eb41abacbdfa905204e897c
44786 F20110115_AABUGX rodriguez_u_Page_33thm.jpg
7b62a28de69c6de4792acf36557cd4a0
cdac6595e2f35da752c4b264a850b2f131c73d84
94078 F20110115_AABTXB rodriguez_u_Page_10.jp2
d9c91597d7fb6f9fd87605d7833585d1
e820f31833d55e28df3f41f4fa3a870029cd6a0c
84603 F20110115_AABTZZ rodriguez_u_Page_34.QC.jpg
d82d880c3ac432b1345e5a3f34346f1d
066836bb08994f72ea712d278c0e5e4c5d10acdd
955195 F20110115_AABUBZ rodriguez_u_Page_19.jp2
70a549030ae5917566c2b7292ff17a60
4fc37221cd25bb858590ebda6d81bfee0e9365c3
66583 F20110115_AABUGY rodriguez_u_Page_10.QC.jpg
7b6beb57fc69825f47414a5cdf8ad9a7
82e0c4b089e7cc25408441245548ecade39fab9d
14339 F20110115_AABTXC rodriguez_u_Page_20.pro
295dfca68512e007558a9fcb570c1241
95aa0c83339255c071ce8f990fdec7a4d070206f
27491 F20110115_AABUEA rodriguez_u_Page_24.pro
04a9eb61c9d4c488b70e13ad18a8102b
e919b2bb390a7b2ba756de1eca76e03edd7ae179
15471 F20110115_AABUGZ rodriguez_u_Page_16thm.jpg
2ae572bd92c7728a7fa236c3948b0212
41182877c9044819c4030cc518f61e708bdca8a1
79376 F20110115_AABTXD rodriguez_u_Page_42.jpg
8dada96f77e1711240000ba9d6e9968e
adebf8fe691e52af08d809bb61073a990d4c688a
4595 F20110115_AABUEB rodriguez_u_Page_26.pro
b2a75a8243734468871a664b93c19831
9794f7a866e8f67a5a70a351a5d349f702a6733d
25484 F20110115_AABTXE rodriguez_u_Page_28.pro
0de3060a70740bd15ba8817d57d3ae1c
3c365129eea66c7243b08529bc3191f7a12850c0
17867 F20110115_AABUEC rodriguez_u_Page_29.pro
b4839b086bca0a9f54abfd4f968713da
fd37710ce4891b3a950960a1feda5b6424fd94bd
13420 F20110115_AABTXF rodriguez_u_Page_41.jp2
6934aba7701503c98cf40f199ae4b603
aa10db2ed36526a43b81db2f20fcea8dbe9ff924
1901 F20110115_AABUED rodriguez_u_Page_31.pro
e9665498044861390424ceb7c3c6454e
0cbb8014e9bb42c15afcfc57c1a5fa9ae1ffc106
771742 F20110115_AABTXG rodriguez_u_Page_15.jp2
dab1e7f763024d0cd5dc2c3d92afd833
dd9191fd301160e23a3f0149dfbc1a22b34a1c89
24985 F20110115_AABUEE rodriguez_u_Page_34.pro
1f6aa625fc187de4ae7b1da7f38b33ed
1fef3c4eb2f37e7f33427254d1d13ba600695ca3
1858 F20110115_AABTXH rodriguez_u_Page_09.txt
f6c5f9924b3524f9b4c4f0986bd4d75e
91811a003e61a3aed3d5b4cc53e1c29955f2e029
10422 F20110115_AABUEF rodriguez_u_Page_35.pro
e25b9af8e1efd669429c979471a0b0d4
1e0e3fea490cc258203b965dd43a43dd7f25124e
437 F20110115_AABTXI rodriguez_u_Page_01.txt
318b29d7b6544eae0bb812f27a68a7b8
1b0937bafd6e9f77b99d954d771eebae9ed01180
46899 F20110115_AABUEG rodriguez_u_Page_37.pro
8c62d25cb0b519f177188a8c3b62a072
28322038ed25961889f3a8176132fbf3d54ee8b1
25061 F20110115_AABTXJ rodriguez_u_Page_33.pro
8a73652a877cb81e63e458ecd8462120
463f5d6dad9586c9e57520b52fb03712ff994b2f
4907 F20110115_AABUEH rodriguez_u_Page_41.pro
0a97e45caf6d56eacee11efdbaed8150
f24cba4aa15e37c106d92c80e67c5d456af7f7ae
43720 F20110115_AABTXK rodriguez_u_Page_24thm.jpg
a97fbd71446d12de4352e227305666fb
1c47e3db65a6ad20a735131779cf285f408fed2b
17829 F20110115_AABUEI rodriguez_u_Page_42.pro
e3fbc91366bdc643e7f8c913da155d53
ede663e1798f86d0dd6905d6a95bc7a82c48d5b4
F20110115_AABTXL rodriguez_u_Page_11.tif
933772ace06d7153b4861aedf7df1587
8956f3978095dc868c3dd20695445bea7f859f47
2330 F20110115_AABUEJ rodriguez_u_Page_02.txt
f426df6b61b0a5ddff14d5f8a0e580e1
826db90e85abcfecdd39105f221733d2070b1376
71527 F20110115_AABTXM rodriguez_u_Page_20.QC.jpg
d2755eed5a644af954336105e5e7ee5e
ffc2982fcc8d43f10f1416bfce0861de380f1335
2053 F20110115_AABUEK rodriguez_u_Page_04.txt
b003cce301391a53f7a9428dc168a127
f515a61f21ba943e2739acfaa56551848481de89
166676 F20110115_AABTXN rodriguez_u_Page_15.jpg
85b076e88ec1f9061cef3881fe561246
d334c1cad830123fc09657dc96cfc3b50670dc69
505 F20110115_AABUEL rodriguez_u_Page_05.txt
42251841f646ce0d95ffc86dc4610ab8
fa5dedf506efb74a259d311a0b83f3149204e7c7
25644 F20110115_AABTXO rodriguez_u_Page_19.pro
a686029b2c91120da8a740a498ca2ac0
2f2a43865e94e0ef6f79a1cf9472a021384dcda1
1552 F20110115_AABUEM rodriguez_u_Page_06.txt
d375a6633fd7ab447efaba3edec50b07
377e1fd46789c223250a06bde95e9a3c86573782
962 F20110115_AABUEN rodriguez_u_Page_13.txt
277a1089a0897ca08f2dea7d6d4ba593
a2535d7a05cb049d583603a37552edfadb753df9
220245 F20110115_AABTXP rodriguez_u_Page_40.jpg
6331ed3ef3517b55ef60d70ee972df48
576c87a73a55f849a04909642df571c2b7961a3c
1245 F20110115_AABUEO rodriguez_u_Page_16.txt
e3be4b7e4b9920cd8b0bdd2ac9e97485
19bb9337ff57b22ddde33285443d747daebccab6
40137 F20110115_AABTXQ rodriguez_u_Page_17.pro
79103f373d60fc867a3350675170073d
27be3b75464e956d88d62eaf0b064409b1629b91
1697 F20110115_AABUEP rodriguez_u_Page_17.txt
30eafa14b2c112ede8f5af66e1261c83
bc18b2b6d8901d538febc7e3e921ecbac56c0e83
F20110115_AABTXR rodriguez_u_Page_03.tif
d0e2ff5f6d6c190da62434ddca85342a
0e6f23fcb5957a0a34377aae99fa03e15ad50d1f
1792 F20110115_AABTXS rodriguez_u_Page_10.txt
f1c34677f58a1149202b7ae1b973aa1d
6541d42f12b0e8f27a9108d6d570b06fac7c9dc0
945 F20110115_AABUEQ rodriguez_u_Page_18.txt
ec4af3c635c086bf7aea217bb1c2a8ec
80b15c3cd82ee6a7795c86f7bc1e318a413ad11a
52655 F20110115_AABTXT rodriguez_u_Page_01.jpg
0145affed538e459532093a12e99ae33
e24fb86bdc7188d7a8ce6df3c8d76c8c6e305c15
1069 F20110115_AABUER rodriguez_u_Page_19.txt
5ecbc95c4db79de717b93a26064568bc
420221a77de8da0d6cb5e859649d76c141f03435
6592 F20110115_AABTXU rodriguez_u_Page_35thm.jpg
cd410725a63e42e98dfdda2bde764936
edbafe559600ca695ea2cff474d60f3d7a8f311a
764 F20110115_AABUES rodriguez_u_Page_20.txt
a7a7f24ed3b85a09534be0a0d3f6c794
1a20c4893e81838852e02993aeb98c326af6cbf8
16194 F20110115_AABTXV rodriguez_u_Page_12thm.jpg
358417bb5db8ab474cdff6d7397621cc
2bd9be875daf130fb5c2ffe06876d7fba5e16d6d
1984 F20110115_AABUET rodriguez_u_Page_21.txt
1148b972ed5214b19768f041acb14069
ae6d681e45cb13a4d22a13ee5077bcefa9978984
F20110115_AABTXW rodriguez_u_Page_08.tif
6f0a9183014177b00c5639163721dff3
e5a3c43e3f89af296857d095a64d49c82bb91e77
1128 F20110115_AABUEU rodriguez_u_Page_24.txt
50cb3edfa5e3f578ec39c32131c20559
cdde0299a908b4482173419bd7855c7f2405e976
821 F20110115_AABTXX rodriguez_u_Page_15.txt
e1954b39dc2c8302522c5f9941b50e88
cae5f47b1d9da992bccd2ce0bcfa65c1c315ac88
267 F20110115_AABUEV rodriguez_u_Page_25.txt
da304616f0d422e14e9d936800f2b00e
84fe5fde8a3a267c50f0181e90b06e1a245c93cb
41763 F20110115_AABTXY rodriguez_u_Page_28thm.jpg
9f701f41ed723672ef8ae3b0510375f1
90451b0633514cfa6aecfe65d2ae78bb41a0f0c2
1491 F20110115_AABUEW rodriguez_u_Page_27.txt
377a9bbb1e028f24b1f86db5700be164
567a9d7a52977ca62c4ab0a8a27ee2b50e55829b
18676 F20110115_AABTXZ rodriguez_u_Page_15.pro
f42743099dc79374246bc919dcc656a9
f32153c3d54204763e0ea5e37f14f42757f0cb6e
1105 F20110115_AABUEX rodriguez_u_Page_28.txt
f003c7e1ca61df25188f48b3dbe3111e
5b017383b41872fdb96fbb1d5f4cd668905ca19f
761 F20110115_AABUEY rodriguez_u_Page_29.txt
914fd14e690f0c57ba496b02dfe7606e
4bea3697c45cebc9cd6b83f1872b07d1363b8d04
804724 F20110115_AABUCA rodriguez_u_Page_20.jp2
038c2faa668c33d6188ce4f59cf62ce8
67d49c0536049b7efad1fa4540919c6b9df3b67c
1569 F20110115_AABUEZ rodriguez_u_Page_30.txt
0153bb55ec6bff2fa83fc65c1d3c80a4
6f17940f346392a835807c8adb64016e48fcce82
103566 F20110115_AABUCB rodriguez_u_Page_21.jp2
57e5643bf6d1869c28cd6c81b459caed
985eb69df6033eb38cb7ddbb197f7271bf1ded0d
968369 F20110115_AABUCC rodriguez_u_Page_22.jp2
a054fe3820a1bd64cac25f58f2e469c1
ac1ec2f0427ae2d5f022c42d8f63ac5d2f0b0354
39900 F20110115_AABUHA rodriguez_u_Page_03.QC.jpg
0832cf46c583a931d36966f6b47e8f60
38aa6c89763834485e1231e9140234c8486e2db7
77793 F20110115_AABUCD rodriguez_u_Page_23.jp2
1811f63bdb1ab5ea9eaa397045f806b7
14242f968b1e873804c82e90d1b6d903e6e76ebe
16143 F20110115_AABUHB rodriguez_u_Page_30thm.jpg
35e5772a89d185124c67da24f5936b9e
6a0cb3a39d015fc5484a5c662e818ddbec460bb3
887550 F20110115_AABUCE rodriguez_u_Page_24.jp2
59caf8906097caf2ec150baa82960e13
c457a4322a409461dcf117d01e897d423c9ab922
30509 F20110115_AABUHC rodriguez_u_Page_42.QC.jpg
1e9b26039c088dde36d8b16a16bf4809
bccd9923c4484afefe2330f1c9e57bbedc06bdb3
621165 F20110115_AABUCF rodriguez_u_Page_25.jp2
343363299cd6573899ae79118c632216
612b58407aa7de3fcb8636b77c0c339d0f47af5f
20790 F20110115_AABUHD rodriguez_u_Page_10thm.jpg
9054dc9ed94a5a89e686f8450373ffc8
e2888c46c11b219481626eba60ba64e6e2eb27a5
669168 F20110115_AABUCG rodriguez_u_Page_26.jp2
086776fecf64e8b3bc23b66c0e2c8ce5
2ee0c11e49f11aec1039fbc7264ac7794e80ead7
29630 F20110115_AABUHE rodriguez_u_Page_07.QC.jpg
914108cc83136f03c752ea102a1feab1
2d1d33730c6fc79b89b46a95a7870fa8ff1f42cf
75236 F20110115_AABUCH rodriguez_u_Page_27.jp2
71798d844596aed531f55fa373ce7191
7db6765dd49843930b565d2f1402caefdb273d48
68346 F20110115_AABUHF rodriguez_u_Page_32.QC.jpg
02e0574d0db04834dbd8a0850907c432
398f50d45322157a803e9497014b5d985e389c4a
607959 F20110115_AABUCI rodriguez_u_Page_29.jp2
88f8016fa500c58f6601a8087a30c74c
69dab0e870d030f53209d31545a85962a405a2fb
49766 F20110115_AABUHG rodriguez_u_Page_30.QC.jpg
80e236e98d122b6095293207c9bfacfb
d0f5e9e40e7249a234710e80da4138a977caff27
1028430 F20110115_AABUCJ rodriguez_u_Page_33.jp2
a8c40603bb863da06b87d9218ecb5801
f129bf39263219c81f09928c481975c16b9bd403
67073 F20110115_AABUHH UFE0008480_00001.xml FULL
b08f6696479e6ae194979a2b498e7bf8
095519ed38be931cb977fc967eb573306c4f37e1
801265 F20110115_AABUCK rodriguez_u_Page_34.jp2
cb55aa9506cc5d580250079b61220937
a77924915b5c05c831a0c22b39e592a183f60d15
70735 F20110115_AABUHI rodriguez_u_Page_37.QC.jpg
d1108f6f30152f1d22ffff8cdff24bfa
74c9f565151119811a250622d6ee0b12d8760444
24642 F20110115_AABUCL rodriguez_u_Page_35.jp2
99951b2ce2df50206d40730043d572a4
f2bad2485cc2d19af6df16d84b726c51f4a9d491
82437 F20110115_AABUCM rodriguez_u_Page_36.jp2
c1329813d733e61e9f5645e7def687a1
3784abe476d63929250b9cd3115e69db4359532d
809899 F20110115_AABUCN rodriguez_u_Page_38.jp2
2a49b930e8fc8f44f284eabfcdeecedd
f19e82798f5d17c09312f39ee0bc3120009de606
100129 F20110115_AABUCO rodriguez_u_Page_39.jp2
eb252bfda95518932fd0a410f6e52ff9
561f7ccc222d8dc68aac33f72398e6dd9b3cb6ab
F20110115_AABUCP rodriguez_u_Page_04.tif
470e21c78a263fe084b1f83a47047487
22c3f6703d49069b366e850f65e7788da13116c5
F20110115_AABUCQ rodriguez_u_Page_05.tif
ddfea2389edcaa71a0c31e7327dc920c
f3e248dcb4e3e0ee38df98b8f943b553a5df95d9
F20110115_AABUCR rodriguez_u_Page_06.tif
43258f534354d2f595fc288ca125d078
01d2ec39326874bee43610c6e01b25dacafdf94c
F20110115_AABUCS rodriguez_u_Page_09.tif
1252e382a00be12299424c51913c7129
b59db2061be0f8a1ad8c668f6d8bfd40dd78955b
F20110115_AABUCT rodriguez_u_Page_12.tif
7481083f97832f28a486026bb2f973e9
a82c4ed5a4c65145a7de5b851b90f6d817d4e12a
F20110115_AABUCU rodriguez_u_Page_13.tif
30de7247d841ee9c69c4388631eeba91
07f0ea1877ff42d7935d8c353bfa5ecc1e95143a
1051984 F20110115_AABTVW rodriguez_u_Page_04.jp2
c205d6d50d53e72ca15c954eb1215295
633f5aceaecce7ba83d477675def15ad2a5d4ec4
F20110115_AABUCV rodriguez_u_Page_14.tif
c3af8fe967a6dbf9930d09fb8e56ce7b
0fae7cf1aab058ca0de5714f16b7aac5b20573d2
794 F20110115_AABTVX rodriguez_u_Page_07.txt
465597f6879714161275a0563e168087
4e679c2ce683cd1ce806c8ac7640cda0f1e6b590
F20110115_AABUCW rodriguez_u_Page_19.tif
261d7d04896e37fff72e5b50c01de62d
09970d9f363168266c07f8a47c7b5f0915cee9cb
42548 F20110115_AABTVY rodriguez_u_Page_02thm.jpg
ab1291467f980c5146ca1efa0065facb
ab239b0d926f4fe036eb5dd5cb5c58c705ceb2b5
F20110115_AABUCX rodriguez_u_Page_21.tif
d796510b086f47a7220e2dfc6a8f14b7
5d33e77617e68ddc8102236456e325450c445f4e
1614 F20110115_AABTVZ rodriguez_u_Page_23.txt
b52d3db7464040f371d698f2324720e0
112910aea02695d555f958cb30a6fb0b0751a321
F20110115_AABUCY rodriguez_u_Page_22.tif
d7cd74b1504e4a99b7c410bad59921f7
2eada556c5f38f85894d0493c563176ce220efc1
130075 F20110115_AABTYA rodriguez_u_Page_30.jpg
86038de2880eda69d8bc578d346d5ca1
7d06839938ee849a7ffd252dd3a678524488f80f
493 F20110115_AABUAA rodriguez_u_Page_35.txt
5d97c5122a6b5df5bfd927ea9b8344a9
0cd9e9d525b27bad937280a9951db7bf527e679f
F20110115_AABUCZ rodriguez_u_Page_23.tif
4b942bd8715f0d0b06db0b9cd701e99a
198bf66c9d4d52d63051a85fa2b53a2e2a2e82f1
75807 F20110115_AABTYB rodriguez_u_Page_28.QC.jpg
f8b1073a093e0c91eec2a4c46aff844b
1cdfe8e549bf1f51fd91b4e5e073fa8fcca3fe79
328 F20110115_AABUAB rodriguez_u_Page_03.txt
45a71dc5e5a74ef3a22343e7b65c2b4a
bed336d377001186756a9fe128a0e342c91427a3
1051982 F20110115_AABTYC rodriguez_u_Page_02.jp2
38390b515374379e52c5bb39caf0b2f7
cadde6e17f97aff59cb2d22f474290945d82bcdf
44924 F20110115_AABUAC rodriguez_u_Page_04thm.jpg
1ac39da303b5a6a95b542248e9aa92b6
aae577f55640e10359054c6cf1e370b4ac7edb21
149 F20110115_AABUFA rodriguez_u_Page_31.txt
c8cdc74aafa7ecebbfedde71a150a204
3654d783637be3bb616c36e1d45d07d426c660ed
233364 F20110115_AABTYD rodriguez_u_Page_04.jpg
731b9504c6b423182c8642438e8a83dd
4646b194e4720c3bf3d78c28becce03e18bff183
713023 F20110115_AABUAD rodriguez_u_Page_31.jp2
20428ed25b56f7479153131f229af86e
f2846dabe7d392cea42d9c2773f0044467844be2
1037 F20110115_AABUFB rodriguez_u_Page_33.txt
cc4873627410c7adfcac12b0279758b2
b63e53007f7b892f9d1719b7e4c250f21dc7bfa3
6847 F20110115_AABTYE rodriguez_u_Page_03.pro
1665ee965ad37ec9759fadaacf3b28d2
f454b25121ab4fae6d56fa0ab79ef162f74d6f28
20827 F20110115_AABUAE rodriguez_u_Page_37thm.jpg
5c1c9800852a9459084dda51dab6ab03
6bb9660c70182975836101d25dac85c93c8c319a
1046 F20110115_AABUFC rodriguez_u_Page_34.txt
c5f3f182656b0c631115b954385b4fe1
6cdfe5454992b2b53017b30a93cfef29cb995e2e
F20110115_AABTYF rodriguez_u_Page_07.tif
818969be20df4f05c4ecdbbb0a9068d6
352bc0a1a1b29e1ff1278626e9b3ac515bdb6262
65090 F20110115_AABUAF rodriguez_u_Page_29.QC.jpg
e285119b39f6e2f6e5d99e3d0875aed1
c552030118fe510dc2cf79463b85d3756a12c25e
1600 F20110115_AABUFD rodriguez_u_Page_36.txt
03bdf39d71ee2b08b1d40ae719b0206a
d7a2339dcee0f16eb21ea52250b9c1be65f80090
46419 F20110115_AABTYG rodriguez_u_Page_39.pro
722d11eb9a3245bd8ec94d5e31cc99f1
29eb2582aff2284a923e68b110f618d3967fbe1c
F20110115_AABUAG rodriguez_u_Page_18.tif
07ac75d944811f2f87693ee75c851a31
9f39e92ed6a5bb8690f4a26546de1c59dcd14a41
1887 F20110115_AABUFE rodriguez_u_Page_37.txt
55de90c8c51d95b0eeddcc5e2a15206d
1c833e7e2d7f480f6169dad88230d082d4fdf9a5
9595 F20110115_AABTYH rodriguez_u_Page_41.QC.jpg
4c7fecacfdbce6bf39d8472859b4e02e
796853f3872aa3790a11daf3f5cdd6fb004049fd
44078 F20110115_AABUAH rodriguez_u_Page_10.pro
8191dbc01851b7452307cfb3ba53e90f
24fc694ca59fd40f932e8f7ad0166ffd926a08b9
1170 F20110115_AABUFF rodriguez_u_Page_38.txt
0f84f1aeceb237602a0002e581cb1712
9b59b876208af90fd81930cab89adae7f6521578
41587 F20110115_AABTYI rodriguez_u_Page_42.jp2
768af3564d11863f96be03f2ec11ab46
7e8d77daf0de923d37a02dab96970ceb15710e8e
F20110115_AABUAI rodriguez_u_Page_20.tif
198f9059ad3c04b3bd4a44bbcb9a4621
5f1f7cf16ced27489ade05372ad20529fcb32564
1892 F20110115_AABUFG rodriguez_u_Page_39.txt
eab1c2dcd874defc43e60de89df8f5dc
b62822155ab3c76e239d1976ab840560ef49841c
74286 F20110115_AABTYJ rodriguez_u_Page_15.QC.jpg
f3d92f60b5a4c70205d087acf4bdc0aa
da22b76715a72c83811e19ed1fdb9648f7b67d75
F20110115_AABUAJ rodriguez_u_Page_32.tif
389bd2ecb7ef3aaf222446c9ae7879f7
02276b449da67c8d90f31f3600471c51d2d915a3
2080 F20110115_AABUFH rodriguez_u_Page_40.txt
0dd9c9533533dd6b23a3b1fe6c6c7336
a8e68399f192fa53c8962e248459d3e03174e7d0
108754 F20110115_AABTYK rodriguez_u_Page_40.jp2
6d9af5be51ab07789df1816871e9ec32
795cd075baf175f51e47bb0e684e98cd29788a23
63032 F20110115_AABUAK rodriguez_u_Page_17.QC.jpg
b3070d1b1cd079635dd6a49e866e5606
73cec8ad29b4b6238d5a401374233b3acf0058a1
729 F20110115_AABUFI rodriguez_u_Page_42.txt
526a2391c2d3c341f864dfa38e8d21de
a6230ca4921c85fba8d1d69cf87257812ced0acd
18045 F20110115_AABTYL rodriguez_u_Page_36thm.jpg
1b5afcf22f0798a5e30b1baf6a3b99c5
821e3f2328caa6508483c78c49718fadb15162fd
147056 F20110115_AABUAL rodriguez_u_Page_06.jpg
cf41a2d62fb7bbb7fc8b3fc407783970
17c64a5bb63151f6f319cf85d64613d66e1d8386
6286 F20110115_AABUFJ rodriguez_u_Page_01thm.jpg
a3e4dea78858e405855ee6d8a2bf4a82
d53ce87330cf733388bc92aeb246101f7260dd17
197726 F20110115_AABTYM rodriguez_u_Page_37.jpg
1c02f8b7cf3cdd28738ab1cd71ff3faf
231eded55d29b7b38db406ecd0e8781ad74ceb38
F20110115_AABUAM rodriguez_u_Page_37.tif
98ca3bd5ff9a046c18fae6286dc9b421
50485f80f547160296b588e7bf1e87863b03d2cf
690047 F20110115_AABUFK rodriguez_u.pdf
c6fcf54ae70223c3909918ba016da01d
8d583cb6bd8069ea9db289a71da2b08b1895dbe3
60236 F20110115_AABTYN rodriguez_u_Page_25.QC.jpg
1cf69eb10aadad3ac1847078a6102386
c328f837c800ae9d45bd7d12f2c19ab330bcd985
73817 F20110115_AABUAN rodriguez_u_Page_30.jp2
f162ce8d16dbdae93c02b65c62919b23
0a0c34292b793ec53060c0f36a19e84f66a09f90
82153 F20110115_AABUFL rodriguez_u_Page_38.QC.jpg
ad09afa5681b7bd47f5ef55f486dbee2
6683b58dd3e7a12506d6e74fbde09c5290ae6350
65815 F20110115_AABTYO rodriguez_u_Page_39.QC.jpg
8e04b16fab8a9e7320c7d125be42a0b9
0e5f622266b171078c2ff14ab7e141f230774bab
142774 F20110115_AABUAO rodriguez_u_Page_32.jpg
9e98a614ca444c45671629bb41bfc84f
fcebadf2f8bfde0737521ea7c97d482c3e818d4d
79465 F20110115_AABUFM rodriguez_u_Page_13.QC.jpg
45d1c3a8f3f4c60bb80f35e55b885152
0685431594f879901a57b26fc4943457a19631e3
1462 F20110115_AABTYP rodriguez_u_Page_12.txt
ffc829f7a89b6a783b192b3e5a2ca623
7e21cdcf13dc849ec5702039d7c35a84f6ea0815
52168 F20110115_AABUAP UFE0008480_00001.mets
1099577a4d8dc18baa15419fe644a5f4
2c0697a3a91b36eb786921f688f2fba11765e7c0
21785 F20110115_AABUFN rodriguez_u_Page_09thm.jpg
08e8c2ed4f7f57da6afbe2bde9464a34
2f88616e0de14d1a47477bda24218b4910316075
84668 F20110115_AABUFO rodriguez_u_Page_24.QC.jpg
b0f700d2d6a5b30c806a27b40520aba1
25dcbd1104f292b4061a1d46842d8e04796e892d
34817 F20110115_AABTYQ rodriguez_u_Page_30.pro
42277aa18ce6567893b1c336b1b7f4d2
08d830d9d48b67bbbf360074e5da9393bb9e23e6
86175 F20110115_AABUFP rodriguez_u_Page_33.QC.jpg
5818e8b9d5e3e0e086bf0fd5c3bc7782
9dfc644d803a9f7af2def70288086aa6dc4e56ba
17665 F20110115_AABTYR rodriguez_u_Page_35.QC.jpg
e0aa056b018b735227bd1710c2d9d509
c0417005bcffd2339ded4aae386105a304c0a70a
87273 F20110115_AABUAS rodriguez_u_Page_05.jpg
aa03f36c1f84d03e945d8fe04b290342
3120407dc7cb6c56a3828d654c39ebb547237afe
53218 F20110115_AABUFQ rodriguez_u_Page_12.QC.jpg
25a41843500e688ebb1328b75b044c9a
010769fe8b34a1dcb91da4e1e0f4fbf492e2f18c
F20110115_AABTYS rodriguez_u_Page_15.tif
e9df77cff177dbda37efee3d862edb8e
9b24a17c3ba41f9b2ba4240da0f039f25742de0e
197229 F20110115_AABUAT rodriguez_u_Page_09.jpg
6dbbd008b8e1cb0eecb137f61c47d0e0
2583570ab359077d0a6037cb940b828d3e1cca88
90579 F20110115_AABUFR rodriguez_u_Page_22.QC.jpg
0cb9260b04eba2151e3f832434b4aa83
2ce88aee3ded31d4311e1962bcdda1cda9efa708
37542 F20110115_AABTYT rodriguez_u_Page_36.pro
11981dc4a0ad0ce872b337348a77bfc7
6656ac9d9021f2ae106e5c585fd1a76143d9965e
188643 F20110115_AABUAU rodriguez_u_Page_10.jpg
ffdc7232c7cabe4f71693fef96bb819f
8fa1adcd82a3c8b679e27fc0117418c9ebeabd81
56030 F20110115_AABUFS rodriguez_u_Page_23.QC.jpg
5736de26727c76c862cd9073d9ebf6d1
5fe8d74556b3827c67706f994ea4093f764fb6f2
72663 F20110115_AABTYU rodriguez_u_Page_09.QC.jpg
d8e2fbed9bb1351e1e9cfc7d46f32367
4a28975d02b425f1d86439c3d1d2b631fb053fce
149315 F20110115_AABUAV rodriguez_u_Page_12.jpg
a6e1615c3370d93224464284d0ced06b
64d19fa901c3d1a7da13131069c66506863d1fa5
42392 F20110115_AABUFT rodriguez_u_Page_13thm.jpg
bb09c01f6902e45b5143ce93a4253261
bfa05cb8d6c42445aa315a8095254e47c1326d5d
46530 F20110115_AABTYV rodriguez_u_Page_22thm.jpg
e4a43c594bba4bb473d2341b12f48ae9
a17ad33989a233b61c9327d2e2570e2cc8846654
185942 F20110115_AABUAW rodriguez_u_Page_13.jpg
83e6373e36cfa2c8c3d3d7583045015e
6e4a110a6708b94d94455a1f0e261f67cfacbf0c
46442 F20110115_AABUFU rodriguez_u_Page_05.QC.jpg
a6c24a698b0a69cc5cdde7f5e75a87b5
7e7f84d5e9617c54b694b84210c4e9c76c11a095
35965 F20110115_AABTYW rodriguez_u_Page_26thm.jpg
4ec7abf81db047b91fff96da67d48189
bbd5aea5ea5bba35da308d63ba6080ddd1e1f089
20882 F20110115_AABUFV rodriguez_u_Page_40thm.jpg
a7054fa1507dc089244a7763806ee813
d5cf99655f361071c1c515b9d74659ee73b97a4b
16888 F20110115_AABTYX rodriguez_u_Page_27thm.jpg
b6bf1b26ecbd009ebb58a7b880a89eeb
1c8d4b447be68f219d58ae4c2c80140e63e5f127
176376 F20110115_AABUAX rodriguez_u_Page_14.jpg
a825d7f214a42a9f4bba8c6854b5aba1
d5a728c5fc9961e5b1af6c4b233c4158c90c89cd
66308 F20110115_AABUFW rodriguez_u_Page_14.QC.jpg
5b1c08672dfc88348964eefe0de3050a
d6bab2b894c0a19a991c4fc50cb40142cad240ed
F20110115_AABTWA rodriguez_u_Page_33.tif
095b88174c4f46e63430484d47c64757
70f151454568140497c8ad07c0fbc1d7850f2b17
F20110115_AABTYY rodriguez_u_Page_01.tif
db03840b562edd793a1b78fbd4cba0ba
112900119282160885a999ec06bbd86b8d506bbd
136593 F20110115_AABUAY rodriguez_u_Page_16.jpg
d78a742070a2e9485bce2285be5cb393
15d8f98725e0297c03f99a5167d50c6b34f4ccb3
61867 F20110115_AABUFX rodriguez_u_Page_08.QC.jpg
cf38148b8a49070f267c794453093002
09f5028448a2d08dc63842a411c726e3c5c126b7
17229 F20110115_AABTWB rodriguez_u_Page_08thm.jpg
61e14bf23935bf00bb1bbf1747970acf
a9b109a54749f968e60e62eb191b9468647e370a
146071 F20110115_AABTYZ rodriguez_u_Page_31.jpg
bffb7f8929bb4bda8140370ef83106e8
b5fa33ee34a8c1d5d86c4b23f63fe663d7687dd3
172594 F20110115_AABUAZ rodriguez_u_Page_17.jpg
bcceeedd9111d4fe9bbc54d20c0ff523
7dac0f2611876d8d46b56a12b6627448c9ccca03
31921 F20110115_AABUFY rodriguez_u_Page_05thm.jpg
75e7264cfc6732049dbee67a8899028f
8f4dd9c6e148c84686420c3d8f2c7c1e48dc6ec2
269 F20110115_AABTWC rodriguez_u_Page_26.txt
47773c89081ae16b577a7167724c3fb3
dc050831ed9a6c37ad07a63ecacd7165492d946e
41313 F20110115_AABUFZ rodriguez_u_Page_15thm.jpg
c891f1a6bb654898df2b9fd0f6346cf0
9e37ec15c59362f98763542a378d513eb52f15a6
210640 F20110115_AABTWD rodriguez_u_Page_02.jpg
8b2223e751182679f88cbfbe774d8483
ea00a484310b48456c6828a2a6e500bff4c5f9c5
F20110115_AABUDA rodriguez_u_Page_25.tif
a54b9b543e2d83e3f3adc84e0964cd6f
8aeb72248e57e5ca331342ce5895ae1d9b090d77
30342 F20110115_AABTWE rodriguez_u_Page_03thm.jpg
e9adb51b8793152f005c5d1986820f20
f270b4097901960d13771fff7c6bc7103250c5e3
F20110115_AABUDB rodriguez_u_Page_26.tif
70a152051eac7c85e1155788cec4d71f
fd4432c4be313818b517828c3c99c0d1e5e45871
38550 F20110115_AABTWF rodriguez_u_Page_29thm.jpg
fccf3516b03e3eba96b1e1c2f00642cf
1f76c2a168ed9c1f23d87f2ff8a69540492d41ea
F20110115_AABUDC rodriguez_u_Page_27.tif
e0842caa6bea07d1280a3da82ef1d863
72b8e7445a4c49f5e9794d1880e76cfa4fac389c
49830 F20110115_AABTWG rodriguez_u_Page_04.pro
3b593e21a71bfd42df6cbce10b203032
ff302283a06883034f81b2ae4745889f11dbaa27
F20110115_AABUDD rodriguez_u_Page_28.tif
ef5acc0c4e94d9cc46fda91fb9b4baef
82d55a4804c66c60c9ce5f6f483475229cce0304
51138 F20110115_AABTWH rodriguez_u_Page_40.pro
d2ebde4253f097e2a0178755f56e708a
a83462b9c43dfac96ce7b0d24c231bef983f8424
F20110115_AABUDE rodriguez_u_Page_29.tif
f59b2852dfaab5c5be574088b436039b
c4ba062adabf74a15292c3f6b2f827b5361b5b7d
49962 F20110115_AABTWI rodriguez_u_Page_21.pro
06d0db2a79dcdbba0bde5cf31673241b
afb3661d14975c1d3b295ef30b87dceb5e32241b
F20110115_AABUDF rodriguez_u_Page_30.tif
06a85a081be6df7141ec9d20c1c01aac
f7247047c1cf661e16af60600f9d1c2df00a9f66
5976 F20110115_AABTWJ rodriguez_u_Page_25.pro
c7d5ea1c81f6e81e589df5ef6cad0f34
f5507a3ac6a09926c53d47d0569de5b3191a31d7
F20110115_AABUDG rodriguez_u_Page_31.tif
583d1b3e55bec6cd1102ce9b68546724
55bb5c28da6647a311ab4dbe7890997b63b2ed95
972687 F20110115_AABTWK rodriguez_u_Page_13.jp2
9e14152afc54837c0b1ffb8cd6d6cc07
702c1b1e4869ce5fbb0c770148e37d380b9dff2b
F20110115_AABUDH rodriguez_u_Page_34.tif
3310c4789bd62c5d796b00323a25f9f4
ccf5428fffb6382a164442028fe9a54817a288b8
F20110115_AABTWL rodriguez_u_Page_24.tif
2a88b9322ebfc3c03eb7cc2d2da0a873
8e66359c1d4f67d2db5d3c47539263a0a3f0ab2f
F20110115_AABUDI rodriguez_u_Page_36.tif
20b0587b625d432f1c656816f409a875
2ab2f53ae1c200a52fcb8b556676e396eb8edd2e
99828 F20110115_AABTWM rodriguez_u_Page_37.jp2
b7caed08905f8513c7194b9580cd16b9
38449030d6f401c751c551c987e1a18e1dbeed6c
F20110115_AABUDJ rodriguez_u_Page_39.tif
4381b64fc54eb46ff058afa4a4ebafa0
de3b31ef2d82c7925dd74023648140237bf935ad
4784 F20110115_AABTWN rodriguez_u_Page_32.pro
cae7c273831c0fe7946d3e17ec4e6c8e
6207f74a8bfee3d20632941780d062982604b8fe
F20110115_AABUDK rodriguez_u_Page_40.tif
1b570aa18afe379e1d360bcd25c23c34
15052723b42ee2d403e4fd59124ab0a6a4da17e2
F20110115_AABUDL rodriguez_u_Page_41.tif
ddb5dc154abcb7a3592db2d12092b002
012ce52a85412d574f16d646b43caab4a412fdc2
74156 F20110115_AABTWO rodriguez_u_Page_21.QC.jpg
890dc1d349c9a01e2ed2ba961d78a286
1c38daf1d9cdcfe487c9b58895ea7ead09bca96d
F20110115_AABUDM rodriguez_u_Page_42.tif
80a008a7e79babd6e62be2c2dd39747e
ed9ae80b32a1b82fa17289a8239597c93196e028
44332 F20110115_AABTWP rodriguez_u_Page_38thm.jpg
dc308e20630ee63ecf381797bba4787b
b0a1c7297071d1d7615ec66e77c386c19fa24ab8
7538 F20110115_AABUDN rodriguez_u_Page_01.pro
4863e3be753f2a0ac1e545e92096c575
2ffc55bdabf9ad3bf4885817e4b129a5957f77fa
54619 F20110115_AABUDO rodriguez_u_Page_02.pro
bfa73b4d82a865918af5009ed77d2c9f
55dfdca89c9208c0855b8826109db2ef71f7f7b4
81890 F20110115_AABTWQ rodriguez_u_Page_07.jpg
71f86ad796eb5c3c527fe26204dc2fbe
f4e3abb71551acc994a87f8077966fd3f8cee1f6
33974 F20110115_AABUDP rodriguez_u_Page_06.pro
0c4ccd74587d063e3335844153f93b83
e99d7153bef16fd3d0ac7bad318ae54cb69d6317
72744 F20110115_AABTWR rodriguez_u_Page_03.jpg
a9f36763c449aeaacef83607edc547af
5bb4c3894e6cbb085459dfc396a6e68dca7cc393
19665 F20110115_AABUDQ rodriguez_u_Page_07.pro
be48eb68d701ddf0ede2b5b41193719c
722f67fa3d82fb2f1aeea42afcc8df542a6a6e21
99548 F20110115_AABTWS rodriguez_u_Page_09.jp2
5e3a570064c11bb229e78135d98b240e
0b1053df86dfbd64408549e319135c96f59dfa96
38421 F20110115_AABUDR rodriguez_u_Page_08.pro
5600f79c36b45ce37250e405a27cad71
e03424668b3690cfea5fe20d09367ec2cf98937c
16264 F20110115_AABTWT rodriguez_u_Page_06thm.jpg
d2f1d4e47821f665bbe6452b8e1ff263
f9aa364f42dd2dea25ef3c017d76c17cc993d130
33617 F20110115_AABUDS rodriguez_u_Page_11.pro
ec5f6dcdefbf438a1fd110ae6db17f15
645a510e202bc329168873bb4507d6b339ecf9fb
276 F20110115_AABTWU rodriguez_u_Page_41.txt
21f7e19070ff6f4cd51ab0345dbadb42
7752c3fb72270955e57479f9684b422c781be306
33777 F20110115_AABUDT rodriguez_u_Page_12.pro
ba66734f38c7f4607959ad9df249256d
8e1071e84be3bc737aca93a4c3d53b83a7ad62af
F20110115_AABTWV rodriguez_u_Page_16.tif
35c5886a03722211ac231521c8764279
1dc1426a60bb6e7ef2e6892ae978bc974e8b58f7
23127 F20110115_AABUDU rodriguez_u_Page_13.pro
07a6efffa4d944f003167cc2511d54f1
324137cea11241e3e9c2cc2472ef2ac38c07cde9
1694 F20110115_AABTWW rodriguez_u_Page_14.txt
6f6add1c6bb4a078bf145b7e884a96ff
b7af6607c6bc7b3e22f756dd449e9065125dec96
41891 F20110115_AABUDV rodriguez_u_Page_14.pro
b6c4bf90f283545b518047630c52c263
a05e08eb8b5ec1c55b05f67940fad6cebc83bb80
165233 F20110115_AABTWX rodriguez_u_Page_08.jpg
14def5145ebfa9f736afd0704cf20ecf
6d00cc0a99c410558ee3b09113fdd1e7e570c252
30846 F20110115_AABUDW rodriguez_u_Page_16.pro
e4fa6bf96753a573fd14a1bfb12ed7cb
336b054abc6df410b1c546a895ee4a591cd4aae5
1537 F20110115_AABTWY rodriguez_u_Page_22.txt
ebcd2a95873ecf3ce27f3bbd973120f1
bf2fff750e44415fe79b5f6fc00dc9ac64943087
22551 F20110115_AABUDX rodriguez_u_Page_18.pro
4cba5a4fd460d54e2756e5a78241d258
656ee3d00440069bad34c122be028c26b7e0323a
72058 F20110115_AABTWZ rodriguez_u_Page_11.jp2
01381f44c615ca5e32c28acbf15d81cc
a86bd4f35c8fd6ab8b5507c43f7a68b788b5b511
35469 F20110115_AABUDY rodriguez_u_Page_22.pro
9a074825c25f0eaae5162b7cc0d4054e
4c96fe648bb5cca268a754c9cfe13d6254119cc1
12562 F20110115_AABTZA rodriguez_u_Page_05.pro
ce39cb2e7a3a97a72269cd4e02a2621c
68f2d9b87aef6c64184eaaea3bf7e286799b178b
193698 F20110115_AABUBA rodriguez_u_Page_19.jpg
00c2f7ba1dcd82784ad5828309d95289
7866b35dfa1798efbad46b6e5307b61655f818e1
35762 F20110115_AABUDZ rodriguez_u_Page_23.pro
4a3da7b44adedd772af87e27f1b12383
82307a20e7c0c416ea314671ecb6b06a039a4529
171293 F20110115_AABTZB rodriguez_u_Page_18.jpg
0a8aa2fa6227fbb2f55c68070f2f112a
04b5d4ae4a45222ef7d83b8b8112b2a698c4737d
152486 F20110115_AABUBB rodriguez_u_Page_20.jpg
0c5d2393bd2907087fafac0943c02cce
c728b8bd4579a63635405fcf3595aa44c0a32f3e
40208 F20110115_AABTZC rodriguez_u_Page_31thm.jpg
b324420c7c5e5c1bfcb6d842af177d99
1ea3936b4db695104ebe4c0b9a53dd8eb4ccacb6
200777 F20110115_AABUBC rodriguez_u_Page_21.jpg
58bece95c70199be67fbf16aa90015f3
d18196d048e434fe6f28492144a05954043e0711
F20110115_AABTZD rodriguez_u_Page_02.tif
56e65d9949ef239441bd88111dfab006
485a6c3b3d838c0dc1ecd8b02824b4e708d17202
202800 F20110115_AABUBD rodriguez_u_Page_22.jpg
64c99568cdb5a0d02c58dd22e63eaea5
4ee55064156b31b86381a7c2e95380da09ec0fd1
15414 F20110115_AABUGA rodriguez_u_Page_01.QC.jpg
d50833471b33d1a151141a73df391544
a1c459e2417172e61913587997b612008bdbffdc
136715 F20110115_AABTZE rodriguez_u_Page_11.jpg
5eb23e42a3a97f4f616767f3dc9a3f93
7f151b5e875fd8f70dee56fe5cad712f943b78ea
189314 F20110115_AABUBE rodriguez_u_Page_24.jpg
bbb5575d6835bc623205675d9b1c033b
36546540ad7c8936a758404d93fa783be2ac5279
49429 F20110115_AABUGB rodriguez_u_Page_11.QC.jpg
93f6f414f0d8501f535cec8845e551a6
4f4f9028f45deb5656cfbf52c9c8eb1fbfb8619b
134276 F20110115_AABTZF rodriguez_u_Page_29.jpg
b40742207a1cca0c1c1e0bec74784500
3db1e8f3e9c796b93dd95b051bc59a4231d7bada
121823 F20110115_AABUBF rodriguez_u_Page_25.jpg
680db7b109c91591c5ab86b5d5320083
e35eead1a96a40a528a2c53a2ffc6752755eb183
10064 F20110115_AABUGC rodriguez_u_Page_42thm.jpg
37f08bd5ea5631a4185ed604be973c8c
b1f1867f74a160ea59c6ad2731041eeaa9668c10
10283 F20110115_AABTZG rodriguez_u_Page_07thm.jpg
f2d3d86fbf2b94fc25fd8932517a1fc4
c0a3b6fc24c316f5fc14fa4a8491ed5aaa480779
128859 F20110115_AABUBG rodriguez_u_Page_26.jpg
63e2e18b00fa16b05a076e650dfdef32
812e01b90c5359e2682cb9e27bd5032d6b97efe5
75650 F20110115_AABUGD rodriguez_u_Page_18.QC.jpg
3d9fd486d0d4790fcb1e869ef9874ed1
a4b63c5291b316ee703096c6b44e4e5290d10559
F20110115_AABTZH rodriguez_u_Page_35.tif
1f7bf326e4f53119ef6bfa01c578f003
9a4eb8e3d1edf8160978740f1a99c51dab903840
152672 F20110115_AABUBH rodriguez_u_Page_27.jpg
e6bbacde17c56480ad21fed5a7bf3792
93df3148b55c53019d66f955ee35b7cb842fb8d3
18022 F20110115_AABUGE rodriguez_u_Page_17thm.jpg
0c86cfdda1b20cd976da0716b9c25e0a
5edeb8d8b287db4c9e9347e8128b447f4874c51c
91767 F20110115_AABTZI rodriguez_u_Page_14.jp2
033350feaed3e7c1ca48299a314601b9
3e88c7830c0bd9f21715a125204a738675387f10
160725 F20110115_AABUBI rodriguez_u_Page_28.jpg
f7993ae0a921bc544e9a5b4fe205bb89
e5a03f9c3c84569be7c992d65a49c82718cbd98a
19337 F20110115_AABUGF rodriguez_u_Page_39thm.jpg
f0b2087b1c7e90fa0b312239e59af8e2
b593b2b3c7ff303c42eb834bea3d004ad108be09
17039 F20110115_AABTZJ rodriguez_u_Page_23thm.jpg
c6d41e4a16a3c4d14fefa81b05dec346
853ceffe3b9496643b2967bf17a4480b928b9d67
195331 F20110115_AABUBJ rodriguez_u_Page_33.jpg
75a5afc1c2821aa09a1361e7cdbaf68e
6e2ff16a09d41c6e2c15b6eeac1c60b4f7852f39
81889 F20110115_AABUGG rodriguez_u_Page_02.QC.jpg
817db70b00be2ec65ef7db93e683ee19
ff11d08f37cae6b16b301f955e26b368c03ae0c2
1639 F20110115_AABTZK rodriguez_u_Page_08.txt
1e84ecbf95616751b15cc2ceccf39fd6
afde2419c59c42266bcd929614e44f5c812d1cf2
182708 F20110115_AABUBK rodriguez_u_Page_34.jpg
325eea212ae9fb4546e86c90968c4f48
52a21f7e82168e7bc082231913ed70cb0b2f95f9
53483 F20110115_AABUGH rodriguez_u_Page_06.QC.jpg
46cdf376a7015a9e987cd50e79eab5e2
be350c20c7b2c912e8352c7c012c1bb15fbeb970
46112 F20110115_AABTZL rodriguez_u_Page_09.pro
7540d62a2e4adeb5f0ff6922b45bed53
7e2b7006526b56832635ed3c430e0f46e5282352
48991 F20110115_AABUBL rodriguez_u_Page_35.jpg
9e1c76708cd6400a56f78bc192aef529
9a38d790959f31917e8d388f73d50511af406a2b
21761 F20110115_AABUGI rodriguez_u_Page_21thm.jpg
148c27989a058334e47fa58ffe9d2325
5eb5f9c3b0d05831964881b87226f870f08a25c4
43772 F20110115_AABTZM rodriguez_u_Page_19thm.jpg
92d82925980b609d14e4a0c00f5a0ab1
992406e95a3ac26dc0e334c17bfba9897a05c73f
168206 F20110115_AABUBM rodriguez_u_Page_36.jpg
69053b86c8322524923e7b0d933d50ce
f5e374d8c10fd133fd7b63a67ccea70051213042
48954 F20110115_AABUGJ rodriguez_u_Page_16.QC.jpg
4500736381669443c6a9a0b53a41da5d
73644a86edba9226544e5c831d887f41c3121984
178276 F20110115_AABUBN rodriguez_u_Page_38.jpg
5d6e3f5b73efb105e68d6809fa1c267f
0f88fdb974e335b4d89ce6ed4b432b5c80a6067b
60721 F20110115_AABUGK rodriguez_u_Page_36.QC.jpg
3d6626694a963263c1de4d090c4ddc50
7e866a07e2e6e90c2667faf6b353bce5528b7bf2
151717 F20110115_AABTZN rodriguez_u_Page_23.jpg
63ec283520ba4087f731202fbfe381f4
f377d8affa058d9fd4769cfc0438e074c06fbcc9
195470 F20110115_AABUBO rodriguez_u_Page_39.jpg
dfa39cada96c6e830cb2e12f081443b1
cf084e2ae44308b071403793378b181d2203a72d
92512 F20110115_AABUGL rodriguez_u_Page_04.QC.jpg
38413c649a46fab35d474e3e31f3f578
97833d06e0d8e0fbd63c1c902c18443027a68b35
710995 F20110115_AABTZO rodriguez_u_Page_28.jp2
5b7ed09fae2f58d39f35ccbe66f4c176
385093fb097bd29cb9397f09b537bcce05e38b9f
31526 F20110115_AABUBP rodriguez_u_Page_41.jpg
bdddf7db337d06c5e506646aa1ce9c25
ab1c274ca6f40c8b48a4743bb27b151f3463fd24
44489 F20110115_AABUGM rodriguez_u_Page_34thm.jpg
da678bf6b1c09c8449215a2f919b78b8
950ffbfe37f823625499df22ac0264664b7252f2
F20110115_AABTZP rodriguez_u_Page_10.tif
e1b86d232401f474c525d35db0fa30f5
89989f22b9c1eebabf12a9ab32d0d3d43b8d75b2
22651 F20110115_AABUBQ rodriguez_u_Page_01.jp2
795a993dbe4f7c7b98016c7611412e1f
d26e16430b4619724af60e005994e142999f141c
36754 F20110115_AABUGN rodriguez_u_Page_25thm.jpg
1826567f61f7fa36c34efa63afc2614e
e6b72e63b92283964b454a18210a0a3bb6787970
258 F20110115_AABTZQ rodriguez_u_Page_32.txt
c2a21362ecf8de08887da0491c746420
d331e297accf024902fdb866739a17317084f7a6
263087 F20110115_AABUBR rodriguez_u_Page_03.jp2
0f47e8ad77cd7ad3712b66d6e36d3a17
d879012a41a0d87cbad416e0ec8cd4c6522ba7b3
70430 F20110115_AABUGO rodriguez_u_Page_40.QC.jpg
06157270fccb3eeb1e74352714e18e85
f94edb84d731e2ce1f943d39f9df8884611f9179



PAGE 1

MINIATURIZATIONOFGROUNDSTATIONFORUNMANNEDAIR VEHICLES By URIELRODRIGUEZ ATHESISPRESENTEDTOTHEGRADUATESCHOOL OFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENT OFTHEREQUIREMENTSFORTHEDEGREEOF MASTEROFSCIENCE UNIVERSITYOFFLORIDA 2004

PAGE 2

TABLEOFCONTENTS page LISTOFFIGURES................................iv ABSTRACT....................................vi 1INTRODUCTION..............................1 1.1Motivation...............................1 1.2PreviousWork.............................2 1.3Challenges...............................3 1.4Overview................................4 2GROUNDSTATION.............................5 2.1PreviousWork.............................5 2.2HandheldGroundStation......................7 2.2.1MotivationforMiniaturization................7 2.2.2SelectionandSpecifcationsoftheHandheldGroundSt ation7 2.2.3Limitations...........................8 3HORIZONESTIMATION..........................10 3.1PreviousWork.............................10 3.2OurApproach.............................11 3.2.1Overview............................11 3.2.2AlgorithmDetails.......................12 3.2.3AlgorithmPerformance....................16 4TESTBEDANDEXPERIMENTALRESULTS..............20 4.1Testbed................................20 4.1.1TestbedSetup.........................20 4.1.2Controller...........................22 4.2Results.................................23 4.2.1VirtualTestbed........................23 4.2.2FlightTesting.........................27 ii

PAGE 3

5FUTUREWORK...............................29 5.1GroundStation............................29 5.2VisionAlgorithm...........................30 REFERENCES...................................32 BIOGRAPHICALSKETCH............................35 iii

PAGE 4

LISTOFFIGURES Figure page 2.1Thefolding-wing"PocketMAV"canbestoredinthecontai nershown.6 2.2Sizecomparisonof12"PowerbooktoZaurus............ ..8 3.1Visualrepresentationofalgorithm................. ...11 3.2ExampleofEttinger'sinitialattemptfailing........ .......12 3.3Theresultofouralgorithmonasampleimageisshownabov e.(a) Theestimatedhorizonisshowninred.(b)Bluepixelswerecl assifedasskywhilegreenpixelswereclassifedasground...... ..13 3.4Theoutputofthebootstrapprocessisshownabove.(a)Bo otstrap imagefromarealrighttest.(b)Distributionofgroundands ky pixelsforthegivenbootstrapimage..................13 3.5Outputofouralgorithmforabankanglegreaterthan45de grees. (a)Estimatedhorizonshowninred.(b)Classifedpixelsand estimatedhorizon.............................15 3.6Correctestimationofhorizon...................... .17 3.7Thehorizonwasproperlyestimatedinbothimageseventh oughboth imageswerenoisy............................18 3.8Theresultsofskippingpixelsareshownabove.(a)Class ifyingevery other5lines.(b)Classifyingeveryother20lines........ ...19 4.1Testbedsetup................................21 4.2Linearestimator.3degreesoferroratworstbetween-45 and45degrees...................................22 4.3Variousimagesfromtherightsimulatorsoftware...... ......24 4.4Controlleroutput..............................25 iv

PAGE 5

4.5Imagesequencefromavirtualenvironmenttestright... .......25 4.6MAVrownintests.............................26 4.7Sampleimagestakenfromtherighttest............... ..27 5.1ThedisparitybetweenCCDandCMOSimagequalityisshown above. (a)ImagetakenwithCMOScamera.(b)ImagetakenwithCCD camera..................................31 v

PAGE 6

AbstractofThesisPresentedtotheGraduateSchool oftheUniversityofFloridainPartialFulllmentofthe RequirementsfortheDegreeofMasterofScience MINIATURIZATIONOFGROUNDSTATIONFORUNMANNEDAIR VEHICLES By UrielRodriguez December2004 Chair:AmauriA.ArroyoMajorDepartment:ElectricalandComputerEngineering Inthisthesis,weseektominiaturizethegroundstationres ponsibleforthe imageprocessingthatisusedtoautonomouslystabilizeunm annedairvehicles (UAVs).Theimageprocessingperformedonthegroundstatio nconsistsofnding thehorizonintheimagestransmittedfromtheUAV'sforward -facingcamera.In conjunctionwithaproportionalderivativecontroller,th eestimatedhorizonisthen usedtostabilizetheUAV.Thus,thefocusofourresearchist oderiveahorizon estimatingalgorithmthatcanbeexecutedinreal-timeonas mallergroundstation. Toachieveourgoals,weproposetouseagroundstationbased onahandheld computer,alsoknownaspersonaldigitalassistant(PDA).T hePDA-basedground stationisconsiderablysmallerthanthepreviousgroundst ationwhichconsistedof anotebookcomputer.However,thelackofaroating-pointun it(FPU)hindersthe developmentofareal-timehorizonestimationalgorithm. vi

PAGE 7

ToaddressthelackofanFPU,weproposetheuseofanon-roati ngpoint intensivehorizonestimationalgorithm.Theoverallappro achofthealgorithmisto ndthetransitionbetweenskyandgroundineachverticalli neoftheimage.The classicationofskyandgroundisachievedbycomparingthe pixelinquestionto thecolormodelsforboththeskyandtheground.Thepixelist henclassiedas skyorgrounddependingonitsnearestcolormodel.Oncethep ositionsofallthe transitionsareknown,linearregressionisperformedonth etransitionpointsto estimatethehorizonintheimage. Ourinitialresultssupporttheoutlinedapproach.Theprop osedalgorithm successfullyestimatesthehorizoninreal-timeenvironme nts. vii

PAGE 8

CHAPTER1 INTRODUCTION 1.1Motivation AmajorfocusoftheUnitedStatesAirForceisthedevelopmen tofunmanned airvehicles(UAVs)thatcanbedeployedinstrategicoperat ionsaswellastactical scenariosincludingtheabilityforsoldiersintheeldtod eploymicroairvehicles (MAVs).AMAVisanUAVwithasmallwingspanthatriesatlowal titudesand slowairspeeds.AMAVconsistsofawingspanandfuselagetha trangesfrom30 inchesdowntosixinchesandoperatesatspeedslessthan25m ilesperhour[1]. TheMAVsoergreatpotentialinboththemilitaryandcivili ansectors. Equippedwithsmallcamerasandtransmitters,MAVscanbeus edtosurveyand monitorareasthataretoofarortoodangeroustosendhumans couts.Possible civilianapplicationsincludemonitoringradiationspill s,forestres,andvolcanic activity.Inthemilitary,MAVsareintendedtoreducetheri skofpersonnelby assistinggroundtroopsinsearchandrescuemissions,trac kingremotemovingtargets,andassessingimmediatebombdamage. However,therearecertainlimitationsthathaverestricte dthewidedeployment ofMAVs.AMAVrequireshighlyskilledpilotstorysuchsmall planes.Thesmall sizeoftheaircraftstendstomaketheMAVsmoresusceptible towindgusts, andthus,hardertostabilizeandcontrol.Thisproblemcanb eremediedbythe developmentofanautonomousMAVthatisself-stabalizable [2].However,the 1

PAGE 9

2 groundstationrequiredtodeployanautonomousMAVisbulky andwoulddisallow theuseofaself-stabilizingMAVincertainsituations.The developmentofa smallergroundstationwouldallowforthedeploymentofaut onomousMAVsin widervarietyofscenarios. 1.2PreviousWork TherehavebeennumeroussuccessfulUAVsimplementedbyvar iouscompanies includingLockheedMartin,NorthropGrumman,andAeroViro nment.Anotable UAVisthePointerandisdevelopedbyAeroVironment[3].The Pointerwas designedasatacticalreconnaissancevehicleformilitary andlawenforcement applications.Anonboardcamerarelayslivevideoimagesto agroundstationthat thepilotcanusetocontroltheUAVremotely.AlthoughthePo interiscurrently beingusedbythemilitary,itslargesizekeepsitfrombeing deployedincertain situations.ForsituationsrequiringasmallerUAV,AeroVi ronmentdevelopedan autonomousMAVnamedtheBlackWidow[4].LikethePointer,t heBlackWidow iscapableofautonomousrightandisequippedwithacameraw hosevideois transmittedtotheground. AttheUniversityofFlorida,theMAVLabhasbecomeverypro cientinthe researchanddevelopmentofsuchsmallaircrafts[5-7].The MAVLabcontinuesto developplaneswithimprovedrightcharacteristics,paylo adcapacity,andstructural integrity.TheLabhasdevelopedplanesthathavewontheInt ernationalMAV Competitionforthepastsixconsecutiveyears.Thecompeti tionjudgesentries basedonthesizeandtheoperationalrangeoftheaircrafts. TheplanesoftheMAV Labwereoriginallycontrolledbyotheshelfremotecontro lairplaneequipment, andtherefore,controlledcompletelybyahumanpilot.Sinc etheoperationalrange

PAGE 10

3 oftheplanewaslimitedbytheabilityofthepilottoeectiv elyfollowthesmall planeintheair,theplaneswerettedwithforward-looking camerasandvideo transmittersthatwouldallowthepilottorytheplaneswell beyondvisualrange. ThecomputervisionresearchattheUniversityofFloridato okadvantage ofthefactthattheMAVsweresendingavideosignaltothegro und.Theinitial visionsystemdevelopedanalyzedtheimagesobtainedfromt heforward-looking camerainordertondthehorizon[8,9,10].Inconjunctionw ithaproportional integralderivative(PID)controller,thehorizonwasused todeterminethenecessaryadjustmentsneededtokeeptheplanelevelwithrespect totheground.Once implementedinreal-time,thevision-basedrightstabilit ysystemcouldkeepaMAV intheairwithoutanyinputfromahumanpilot.Theadvantage ofanautonomous MAVisthatitallowsunexperiencedpilotstorythemsinceth eMAVsarecapable ofself-stabilization. 1.3Challenges Initially,theimageprocessing,requiredtoestimatetheh orizon,wasperformed onthegroundusingadesktopcomputer.Astechnologyimprov ed,thevision algorithmwasportedtoaground-basednotebookcomputer.H owever,ifground troopsaretodeployMAVsonthebattleeld,thenitisessent ialthattheground stationbeascompactaspossible. Anumberofformidablechallengeshavebeenencounteredwhe nminiaturizing thegroundstation.Thesechallengesresultfromfeaturest hatareunavailable oncurrenthandheldcomputers.Thesemissingfeaturesincl udethelackafast input/output(I/O)portandthelackofaroating-pointunit (FPU).

PAGE 11

4 Inpreviousgroundstations,theFireWireport(IEEE1394)w asusedin conjunctionwithaframegrabbertoimporttheincomingvide ostreamtothe groundstation.However,handheldcomputersarecurrently lackingaFireWireport oranequivalentlyfastI/OportlikethesecondgenerationU niversalSerialBus (USB2).Theproposedapproachistousethecompactrashcard slottoimportthe videostreamviaawirelessInternetconnection. Sinceroating-pointinstructionstakeaboutanorderofmag nitudelongerto executeonaprocessorwithoutanFPU[11],thelackofanFPUw illlimitthe abilityofthevisionsystem.Anon-roatingpointintensive horizonestimation algorithmisproposed. 1.4Overview Inthisthesis,werstintroducethemotivationandchallen gesforminiaturizingthegroundstation.Chapter2discussestheneedforasma llergroundstation andimplicationsofmigratingthegroundstationtoahandhe ldcomputer.Chapter 3describesthehorizondetectionalgorithmdesignedtoove rcometheshortcomings ofthehandheldcomputer.Chapter4describesthetestbedse tupandillustratesthe experimentalresultsofautonomousrightsinasimulatoras wellasinrealrights. Finally,Chapter5oerssomeideasforfuturework.

PAGE 12

CHAPTER2 GROUNDSTATION InthisChapter,werstdescribetheevolutionoftheground stationandthe motivationtoshrinkthegroundstationevenfurther.Wethe nproposetheuseofa handheldcomputerasthenextgroundstation.Finally,weex aminethelimitations encounteredwhenusingahandheldcomputerasthegroundsta tion. 2.1PreviousWork Ourultimategoalistoperformthehorizonestimation,used tostabilizethe aircraft,on-boardtheMAV[12,13].On-boardimageprocess ingwouldeliminate thetransmissionnoiseproblemsandframedropoutscausedb ytransmitting thevideotothegroundforprocessing.However,becauseoft hecomputational demandsofthehorizonestimationandthepayloadlimitatio nsoftheMAV,the calculationspresentlyneedtobeperformedontheground. BoththePointerandtheBlackWidowuseagroundstationthat pilotsuseto controltheUAVs.ThePointergroundstationisthesizeofab riefcase{7"x11" x16"[3].TheBlackWidowwentthroughseveraliterationsof itsgroundstation [4].Itstartedasacollectionofo-the-shelfcomponentst hathadtobeassembled ontheeldandevolvedintothesizeofa15-lbbriefcase.Eve ntually,theBlack Widow'sgroundstationshrunktothesizeacompactPelicanc ase. 5

PAGE 13

6 Figure2.1:Thefolding-wing"PocketMAV"canbestoredinth econtainershown. AttheUniversityofFlorida,therstgroundstationconsis tedofa900MHz desktopcomputerwithaframegrabbercardrunningtheMandr akeLinuxoperatingsystem[8].Ascomputertechnologyimprovedovertheyea rs,thegroundstation wasfurtherreducedfromadesktopcomputertoa12"ApplePow erbooknotebook computerrunningat1GHz[14].SwitchingtoanApplePowerbo okremovedthe needforaframegrabbercardsincethenotebookisequippedw ithaFireWire 400(IEEE1394a)portandaQuickTimeAPIthatprovidesaneas yinterfaceto captureimagesfromtheFireWireport.Anotheradvantageof thisApplecomputer isitsoperatingsystem{OSX.OSXisUNIX-basedoperatingsy stem,andthus, allowsforeasymigrationofexistingcodefromaLinuxenvir onmenttotheOSX platform.

PAGE 14

7 2.2HandheldGroundStation 2.2.1MotivationforMiniaturization In2003,theMechanicalandAerospaceEngineeringDepartme ntatthe UniversityofFloridareceivedagrantfromtheU.S.Armytha twouldrequirethe developmentofaplanewithasmallstoragefootprint.Thisg rantwasforthe developmentofaplaneandthegroundstationthatcouldtin thepocketofa soldier'sbattledressuniform(BDU).TheBDU'spocketdime nsionsare8"x8"x 2".TheMAVLabproducedafolding-wingairplanethatwhenfo lded,cantin acontainerthatis8"x4"x2".Theremaining8"x4"x2"ofunoc cupiedspace intheBDUistoosmalltoaccommodatethe12"Powerbook.Ther efore,asmaller groundstationneedstobedeveloped.BoththeMAVandthecon tainerareshown inFigure2.1.2.2.2SelectionandSpecicationsoftheHandheldGroundSt ation TheSharpZaurusSL-5600isthehandheldcomputer,alsoknow nasapersonal digitalassistant(PDA),selectedtobecomethenextground station.TheZaurus specicationsareshowninTable2.1.Themainadvantageoft heZaurusisits EmbeddedLinuxOperatingSystem.ALinux-basedoperatings ystemisan advantagebecauseonceagain,anypreviouslydesignedcode canbecross-compiled forthisdevice.Italsoprovidesafamiliarsetofdevelopme nttools,including,the GNUDevelopmentTools. Amoreobviousadvantageofthehandheldisitssize.Measuri ng0.9"x 5.4"x2.9",theZaurusismorethan12timessmallerinvolume thanthe12" Powerbook(1.35"x12.7"x10.2").At7.1oz,theZaurusweigh s13timeslessthan

PAGE 15

8 Figure2.2:Sizecomparisonof12"PowerbooktoZaurus. thePowerbook(5.9lbs).Figure2.2illustratesthephysica ldierencebetweenthe ZaurusandthePowerbook.2.2.3Limitations TheZaurusaddstwomajorlimitations.EventhoughtheZauru sutilizes Intel'sXScaleprocessorrunningat400MHz,itislackingaF loating-PointUnit (FPU).Sinceroating-pointoperationsmustbeperformedin softwareonthe Table2.1:SpecicationsoftheSharpZaurusSL-5600. CPUIntelXScale(PXA255,400MHz) PlatformLinux2.4(Embedix) Display3.5"RerectiveTFTColorDisplaywith240x320Resol ution Memory32MBSDRAM/64MBFlashROM I/OCompactFlashSlot,SDCardSlot,SerialPort

PAGE 16

9 Zaurus,roating-pointoperationsplaceahugeburdenonthe processor,takingup toanorderofmagnitudelongertoexecutethanintegeropera tions[11].Therefore, thealgorithmperformedontheZaurusshouldbedesignedtoa voidusingroatingpointcalculations. AnotherlimitationoftheZaurusSL-5600isthelackofFireW ireorUSBports. ThiseliminatesthepossibilityofusingaFireWireorUSBfr amegrabbertoimport thevideofeedfromtheplane.An802.11bWirelessCompactFl ashCardonthe ZaurusandahostcomputerwithaFireWireportandnetworkac cessareusedto importtheimagesintotheZaurus. BrighamYoungUniversity(BYU)hassuccessfullyusedPDAsa saninterface toUAVs[15].BYUemploysasimilarmethodofusingahostcomp utertocommunicatetothePDAviaan802.11blink.Itisimportanttonotet hatBYUhasonly usedthePDAasaninterfacetotheMAVwhilestillusingtheho stcomputerto handlethedataprocessingandcommunicationtotheUAV. InthenextChapter,wewilldiscussthepreviousattemptsto detectthe horizonanddescribeanalgorithmthatwillsuccessfullyes timatethehorizonona PDAinreal-time.

PAGE 17

CHAPTER3 HORIZONESTIMATION InthisChapter,wediscussthepreviousworkdonetoestimat ethehorizon. Wealsoproposeanewapproachthatwillallowustoestimatet hehorizonona handheldcomputer.Finally,theperformanceofthenewappr oachisdiscussed. 3.1PreviousWork Intheinitialattempttoestimatethehorizon,themainassu mptionwasthat theskywouldberepresentedbyhigh-intensity(light)imag eregionwhilethe groundwouldberepresentedbythelow-intensity(dark)ima geregion[8].Locating theboundarybetweentheskyandthegroundwouldresultinth ehorizon.The basicideabehindthealgorithmwastotastepfunctiontoth eintensityvalues ofeachverticallineintheimageasshowninFigure3.1.Once thepositionofthe best-ttransitionsforeachverticallineintheimagearek nown,alinearregression wasperformedonthesetof(x,y)positions.Theresultingli newouldestimate thehorizon.Althoughthisapproachworkedwhenryingovera uniformforeston asunnyday,itresultedinlargeerrorsinthehorizonestima teswhenryingover groundobjectsofhighintensity.Anexampleofthiserroris showninFigure3.2. Sinceonlymarginalresultswereobtainedfromtherstalgo rithm,adierent approachwastakentocorrectlyestimatethehorizon.Inste adofcombiningthe colorinformationintointensity,andthus,losingtwo-thi rdsoftheinformationavailableintheprocess,bothskyandgroundweremodeledasastat isticaldistribution 10

PAGE 18

11 Figure3.1:Visualrepresentationofalgorithm. incolorspace[16].Thetaskwasthentondthesetofpointsw ithintheimage thatwouldhavethehighestlikelihoodofttingthegivendi stributions.Thisis accomplishedbyperformingasearchthroughthepotentials etofalllinesinthe imagespaceandndingthelinethatbesttthedistribution s. Althoughthecolormodel-basedhorizondetectionalgorith mwasdemonstrated toworkat30Hzinreal-timewithover99.9%correcthorizoni dentication,itwill notrunanywherecloseto30Hzinreal-timeontheZaurusbeca useofitsreliance onroating-pointcalculations[16]. 3.2OurApproach 3.2.1Overview InordertostabilizetheMAVwiththeZaurus,arobust,yetno n-roating-point intensivehorizondetectionalgorithmmustbedeveloped.T henewalgorithm

PAGE 19

12 Figure3.2:ExampleofEttinger'sinitialattemptfailing. borrowsfrombothofthealgorithmsdiscussedabovetocreat ealightweight,yet eectivehorizondetectionsystem. Theoverallapproachofthealgorithmistondthetransitio nbetweensky andgroundineachverticalline.Pixelsareclassiedaccor dingtoitsnearestsky orgroundcolormodel.Onceallthe(x,y)positionsofthetra nsitionsforeach verticallineareknown,alinearregressionisperformedto estimatethehorizonas illustratedinFigure3.3.3.2.2AlgorithmDetails Therststepintheestimationofthehorizonistobeabletoc lassifyeach pixelintheimageaseitherskyorground.Acolor-basedmode lisbuiltofboth theskyandtheground.Notsurprisingly,itturnsoutthatth eblueandgreen channelsofanimagecontainingahorizonproduceadistinct distributionofthe skyandthegroundpixels.Abootstrapprocedureisperforme dbeforeeachaircraft

PAGE 20

13 (a) (b) Figure3.3:Theresultofouralgorithmonasampleimageissh ownabove.(a)The estimatedhorizonisshowninred.(b)Bluepixelswereclass iedasskywhilegreen pixelswereclassiedasground. (a) (b) Figure3.4:Theoutputofthebootstrapprocessisshownabov e.(a)Bootstrapimagefromarealrighttest.(b)Distributionofgroundandsky pixelsforthegiven bootstrapimage.

PAGE 21

14 launchtobuildthecolormodel.Thisisaccomplishedbypoin tingtheplane's cameratowardsthehorizonandassumingthatthetopone-thi rdoftheimageis skyandthebottomthirdoftheimageisground.Figure3.4sho wsthecolormodel distributionresultsofthebootstrapsequence. Oncethecolormodelsareacquired,theimageistraversedin thevertical direction.Aseachcolumnistraversed,eachpixelisclassi edasskyorground dependingonitsnearestcolormodel.Thestartingandendin glocationsofthe largestconsecutivesetofskyandgroundpixelsineachcolu mnarealsorecorded. Next,thetransitionbetweenskyandgroundisdeterminedby theskypixelfrom thelargestconsecutivesetofskypixelsthatisclosesttot helargestconsecutive setofgroundpixels.Thisisbasedontheassumptionthatgro undpixelsaremore likelytobeclassiedasskypixelsthanskypixelsbeingcla ssiedasgroundpixels. Thisalgorithmdoesnotassumethateveryverticallinewill includeatransitionfromskytoground,aswouldbethecaseinanimagewithar ollanglegreater than45degrees.Therefore,thealgorithmmustdeterminewh enatransitionhas notoccurredanddiscardthetransitionpointgeneratedbyt hatline.Agoodtransitionmustmeetthefollowingtworequirements:(1)Theend ofthelargestsetof skypixelsthatisclosesttotheedgeoftheimagemustbeclos eenoughtotheedge oftheimage,and(2)theotheredgewhichisnearesttothemid dleoftheimage mustbefarenoughfromtheedge. Itshouldbeevidentthattherewillbecasesinwhichonlyafe wgoodtransitionswillbefoundbythealgorithm.Thisagainisthecasein animagewitharoll anglegreaterthan45degrees.Onewaytoapproachthisprobl emistocalculatean initialhorizonestimate,theniftherollangleisgreatert han45degrees,repeatthe

PAGE 22

15 (a) (b) Figure3.5:Outputofouralgorithmforabankanglegreatert han45degrees.(a) Estimatedhorizonshowninred.(b)Classiedpixelsandest imatedhorizon. processusingthehorizontallinesinordertoincreasethea ccuracyofthehorizon. However,thisapproachwouldrequirethelinearregression tobeperformedtwice, oncefortheverticallinesandonceforthehorizonlines.Th elinearregression calculationareroating-pointintensive,andtherefore,w ouldhurttheoverallperformance.Moreover,itmightbepossiblenottobeabletoest imateahorizonfrom theverticallinesiftherollangleiscloseto90degrees.Th ealternateapproach usedtoeliminatetheextraroating-pointcalculationsist okeeptrackofthenumberofgoodtransitions.Basedonthenumberofgoodtransiti onsfoundandthe sizeoftheimage,itispossibletodetermineiftheprocessn eedstobeperformedon thehorizonlines.Theresultofahorizontalsweepisshowni nFigure3.5. Oncethepixelshavebeenclassiedandtheminimumnumberof good transitionpointshavebeenidentied,linearregressioni sperformedonthegood transitionstoestimatethehorizon.Linearregressionout putstheslope, m ,and they-intercept, b ,ofthelinethatcorrespondstotheestimatedhorizonandis

PAGE 23

16 calculatedbyusingthefollowingformulas: m = n P ( xy ) P x P y n P ( x 2 ) ( P x ) 2 ; (3.1) b = P y P ( x 2 ) P x P xy n P ( x 2 ) ( P x ) 2 ; (3.2) where x and y representthe(x,y)coordinatesofgoodtransitionsidenti edinthe image.Therollangle(indegrees), ,canbecalculatedbytheinversetangentof theslopeofthehorizonline: = atan ( m ) 180 =: (3.3) Thisalgorithmalsoallowsawaytoeasilydeterminetheorie ntationof theaircraftatalltimes.Whileperformingverticalscans, ifthelargestsetof consecutiveskypixelsisabovethelargestsetofconsecuti vegroundpixels,then theplaneisryingnormal.Ifthelargestsetofconsecutives kypixelsisbelow thelargestsetofconsecutivegroundpixels,thentheplane isryingupsidedown. Similarly,itispossibletodeterminethelocationofthesk yrelativetotheground whileperforminghorizontalscanswhentheplaneisinastee prollangle. 3.2.3AlgorithmPerformance Theperformanceofouralgorithmismorerobustthantherst algorithm discussedinthebeginningofthisChapter.Aspreviouslysh owninFigure3.2, theintensity-basedalgorithmfailedtocorrectlyestimat ethehorizonwherehighintensityobjectswerepresentontheground.Figure3.6sho wsthecorrectly estimatedhorizonachievedbyusingouralgorithmonthesam eimage.Our

PAGE 24

17 Figure3.6:Correctestimationofhorizon. algorithmcanalsocorrectlyestimatethehorizoninnoisyi magesasshownin Figure3.7. OuralgorithmalsoexecutesmuchfasterinaCPUwithoutanFP Uthanthe previousalgorithmssincethenewapproachusessignicant lylessroating-point operations.Performancecanbefurtherenhancedbyskippin glines.Ifeveryother lineisskipped,theverticalscantimewouldbedecreasedby one-half.Ifeverythird verticallineisanalyzed,thentheverticalscantimewould bedecreasedbyathird, andsoforth.Thedierencebetweentheoutputproducedbyan alyzingeachline andtheoutputproducedbyskippinglinesisnegligible.The algorithmresultsof skippingpixelsisshowninFigure3.8.Theamountofpixelst hatareskippable dependsonthesizeoftheimage.Foralargerimage,moreline scanbeskipped. InthenextChapter,wewilldiscussthetestbedsetupusedto testthealgorithminareal-timeenvironment.Thissetupwillinclude rightsinavirtual

PAGE 25

18 Figure3.7:Thehorizonwasproperlyestimatedinbothimage seventhoughboth imageswerenoisy.environmentaswellasrealtestrights.Finally,theresult sofboththevirtualand realtestrightswillbeshown.

PAGE 26

19 (a) (b) Figure3.8:Theresultsofskippingpixelsareshownabove.( a)Classifyingevery other5lines.(b)Classifyingeveryother20lines.

PAGE 27

CHAPTER4 TESTBEDANDEXPERIMENTALRESULTS InthisChapter,wediscussthetestbedsetupusedtotestthe visionalgorithm inareal-timeenvironment.Weshowhowavirtualenvironmen tisusedtoverify thealgorithmbeforearealtestrighttakesplace.Finally, theresultsoftheright testsarediscussed. 4.1Testbed 4.1.1TestbedSetup ApictorialdiagramofthetestbedsetupisshowninFigure4. 1.Thevideo signalistransmittedfromtheairborneMAVtothegroundvia a2.4GHzRF transmitter.ThesignalisthenfedintoalaptopviaaFireWi reframebuerthat down-samplestheimageto80x60resolution.Thelaptopthen transmitsthedownsampledimagetotheZaurusviaan802.11blinkforimageproc essing.Afterthe horizonisestimatedontheZaurus,thedesiredservopositi onsaretransmitted viaanRS-232linktoan8-bitmicro-controllerthatconvert sthemtoaservo train-pulsethatisthenfedtotheRCcontroller.Whenthetr ainerswitchonthe controllerisengaged,thecontrollertransmitsthetrainpulsegeneratedfromthe desiredservopositions.Whenthetrainerswitchisnotenga ged,thenahumanpilot hascontroloftheplane.Thisisbenecialforrighttesting incasesomethinggoes wrong,andthehumanhastoretakecontroloftheplanetoavoi dapotentialcrash. 20

PAGE 28

21 Figure4.1:Testbedsetup. Theimageneedstobedown-sampledto80x60resolutionorsma ller,not becauseoftheperformanceofthealgorithm,butbecauseoft hebottleneckcaused bythebandwidthofthe802.11bconnection.An8-bitdepth,u ncompressed RGBimagearrivingat30Hzwouldneedabandwidthof3.4Mbps( 8bits 3 colorchannels 80width 60height 30timespersecond).Althoughthe 802.11bprotocolhasarawrateoftoup11Mbps,itsactualthr oughputisabout halfofthat[17].Also,assignalstrengthandqualitydecre ases,thethroughput diminishes.Iftheimagesizeisincreasedto120x80,thenth erequiredbandwidth wouldbe6.7Mbpswhichismorethanthe5.5Mbpsactualthroug hputbandwidth of802.11b.However,thereisapositiveside-eecttodownsamplingtheimage. Thedown-samplingactsasameanlter[18],sothereareless outliersinthecolor distribution.

PAGE 29

22 Figure4.2:Linearestimator.3degreesoferroratworstbet ween-45and45degrees.4.1.2Controller Inthepast,complicatedcontrollershavebeenusedtostabi lizetheplane [19,20];however,asimplerapproachistakenintheexperim entsperformed.A simpleproportionalderivative(PD)controllerisusedtoc orrecttherollofthe plane[21].Equation4.1isusedtocalculatethedesiredser voposition, d ,where K p and K d aretheproportionalandderivativegains,respectively,a nd n isthe neutralservoposition.Therollangle, ,calculationcanbesimpliedfroma roating-pointinversetangent(4.2)toanintegermultipli cation(4.3)viaalinear

PAGE 30

23 estimator. d = n K p K d ; (4.1) =arctan( m ) 180 =; (4.2) 0 m 48 : (4.3) Thelinearestimatorfortherollangleworksbestforangles between-45and 45degreeswheretheerrorisless4degreesasshowninFigure 4.2.Anon-linear estimatorcaneasilybebuilttoproducesmallerrorsforall possiblerollangles. Thefactthatthelinearestimatorcausesalargeerroratang lesgreaterthan 45 degreesisnotaproblemsincetheeectthattheerrorcauses isalargercorrection rate. 4.2Results 4.2.1VirtualTestbed Avirtualenvironmentisusedtotestthehardware,software ,andalgorithm beforeanactualtestright[22].Thevirtualtestbedisbase donano-the-shelf remotecontrolairplanesimulationpackage.Thissetupuse sthesameinterfaceto thegroundstationthattheMAVuses,andthus,allowsforsea mlessinterchanging betweenthetwo.Insteadoftransmittingtheplane'svideot othegroundstation, thesimulator'svideooutputisfedintothegroundstation. Similarly,theRC controller'soutputbecomestheinputtotheairplanesimul ator.Thesimulation packageoersadiversesetofscenerythatallowsfortothet estingofhardware, horizonestimation,andthecontrollerwithouttheneedtor yaMAV.Figure4.3

PAGE 31

24 Figure4.3:Variousimagesfromtherightsimulatorsoftwar e.

PAGE 32

25 Figure4.4:Controlleroutput. Figure4.5:Imagesequencefromavirtualenvironmenttestr ight.

PAGE 33

26 Figure4.6:MAVrownintests. illustratessomeoftheenvironmentsavailableinthesimul ationpackageaswellas theresultsfromthehorizonestimator. AlthoughtheinitialresultsasshowninFigure4.3aregood, alaghasbeen encounteredthatrenderedthesystemunusableinareal-tim eenvironment.Thelag isintroducedbythetimerequiredtosendtheimagefromtheh ostcomputertothe Zaurusvia802.11b.Bythetimethattherstimageistransfe redtotheZaurus, thenextimageisreadytobetransfered,andthus,leavingvi rtuallynotimeforthe imageprocessing.Thisproblemisremediedbyonlytransfer ringeveryotherframe totheZaurus.Thistechniquediminishestheeectofthelag andallowstheentire processtoexecuteinreal-timeat15Hz. Thecontrollerperformswellwhilerunninginreal-timeat1 5Hz.Theservo commandsenttothesimulationasproducedbythecontroller tostabilizetheplane

PAGE 34

27 Figure4.7:Sampleimagestakenfromtherighttest. isshowinFigure4.4.Figure4.5showsasequenceofimageswh eretheplaneis stabilizingthehorizon.4.2.2FlightTesting Afternalizingthesystemparametersinthevirtualenviro nment,theground stationisreadyforarealtestright.TheUAVrownforthisex perimentisshown inFigure4.6.Onitsmaidenright,thegroundstationproduc edexcellentresults. ThePDA-basedgroundstationcorrectlyestimatedthehoriz onandsuccessfully stabilizedtheplaneinreal-time.Thecontrollergainsuse dinthevirtualtestbed workedontherealplanewithoutrequiringanyadditionalad justments.Theimages showninFigure4.7aretheresultofthehorizonestimatoron imagestakenfrom thetestright. Anunforeseenadvantagewasencounteredwhilerighttestin g.Inthepast,the pilothasbeenconstrainedtoryingtheplanejustafewfeeta wayfromtheground

PAGE 35

28 stationbecausethepilot'sRCcontrollerwasusedtosendco mmandstotheplane. Now,thepilotisfreetoroamasheisryingtheplanebecauseo nlythePDAis attachedtohisRCcontroller.Theotherpartsofthegrounds tationwhichinclude theantenna,receiver,framegrabber,andlaptopcanbeloca tedawayfromthepilot sincethedataistransmittedtothePDAwirelessly.

PAGE 36

CHAPTER5 FUTUREWORK Thesystemsdevelopedinthisworkhaveonlytakeninitialst epstowarda small,stand-alonegroundstation.ThisChapterwilldiscu ssimprovementstoboth thegroundstationandthevisionalgorithmthatwillallowu stomeetthosegoals. 5.1GroundStation Astechnologyimproves,thepotentialandimportanceofthe PDAasastandalonegroundstationforaUAVwillincrease.Therearealrea dybetterPDAsin themarketsportinga624MHzCPUand64MBofSDRAM.Thatisalr eadya64% increaseinprocessingpoweranddoubletheamountofRAM.An ewtransrective TFTVGAdisplaywith640x480resolutionisahugeimprovemen tovertheold 320x240resolutionrerectiveTFTdisplaywhichhadpoorvis ibilityindirect sunlight.ThenewPDAsalsocarrya2Dand3Dvideoaccelerato rwith16MBof videomemory.Thevideoacceleratorwillgreatlydecreaset heamountofCPU processingpowerrequiredtorepaintthescreenwhendispla yingtheincoming imagestothePDAdisplay. AlthoughsomePDAshavethecapabilitytoactasaUSBhost,an dthus,allow thepossibleuseofaUSBframegrabberthatwouldeliminatet heneedforlaptop computer,thelimitationliesintheinabilitytoeasilycus tomizetheoperating system.CompleteembeddedLinuxdistributions,suchasOpe nZaurus,arenot yeteasytocongureandcustomize,butwillbeavailableint henearfuture.This 29

PAGE 37

30 abilitywouldnotonlybebenecialtotheoverallperforman ceofthesystemby trimmingneedlessfeaturesfromthekernelandotherunnece ssaryapplications,but itwouldalsobeneededtocompileVideo4Linuxandotherfeat uresintothekernel whicharerequiredbymostframegrabbermodules. Ideally,FireWiresupportwouldbeaddedtoPDAs.Unfortuna tely,thereisnot muchdemandforsuchafeature.TheCPUswouldprobablyneedt oreachspeeds ofatleast1GHzbeforesuchafastdatabuscanbefullyutiliz ed.Afterall,there's noneedtotransferinformationsoquicklyintoasystem,ift heprocessorcannot keepupwithit.AFloating-PointUnitwouldalsobeawelcome deditiontothe PDA,butagain,thereisnotmuchdemandforitinthecurrentm arket.Hopefully, asPDAsbecomemoreandmorepowerful,theywillstartreplac inglaptops,and thus,intheprocessberetrottedwithsomeofthefeaturest hattheycurrently lack. 5.2VisionAlgorithm Althoughthevisionalgorithmwassuccessfulinestimating thehorizon,steps canbetakentoimproveitsresults.Theperformanceofthevi sionalgorithmis directlycorrelatedtoitsabilitytoclassifyskyandgroun dpixels.Betterquality imageswouldleadtobettercolormodelswhichinturnwouldi ncreasethepercentageofcorrectlyidentiedskyandgroundpixels.ACCD( chargedcoupled device)cameraontheaircraftwouldproducemorevivid,low er-noiseimagesthan thecurrentCMOS(complementarymetaloxidesemiconductor )camerasusedfor therighttests.ThelightsensitivityofCMOScamerasismuc hlowerthanCCD cameras.Figure5.1showsthedramaticdierencebetweenan imagecapturedwith aCMOScameraandanimagecapturedwithaCCDcameraonthesam eday.A

PAGE 38

31 (a) (b) Figure5.1:ThedisparitybetweenCCDandCMOSimagequality isshownabove. (a)ImagetakenwithCMOScamera.(b)ImagetakenwithCCDcam era. dynamiccolormodelcouldalsoboostclassicationresults aslightingconditions changethroughouttheday. Moretraditionalimageprocessingtechnicslikeerosionan ddilationcanreduce unwantednoiseormisclassicationsinanimage.Also,ablo b-ndingtechnicto ndandeliminatesmallsetsofconnectedpixelscanbeuseds imilarlytoerosion anddilationtoimprovetheoverallqualityoftheimageafte rclassication. Thecontrollercouldalsobegreatlyimproved.Amodel-base dcontrollerusing statefeedbackfromthevisionsystemwouldbeatremendousi mprovementover thecurrentproportional-derivativecontrollerapproach .Thecontrollercouldalso implementjoystickcontrolfromthePDAtousetodirectlyma nipulatetheplane. Alongwithacompactrashmodem,thiswouldeliminatethenee dforthe8-bit micro-controllerandtheRCcontroller.

PAGE 39

REFERENCES [1]J.M.McM.andCol.M.S.Francis,\MicroAirVehicles-Tow ardaNew DimensioninFlight,"WorldWideWeb, http://www.darpa.mil/tto/mav/ mav_auvsi.html ,lastaccessedonNovember15,2004. [2]J.W.Grzywna,J.Plew,M.C.Nechyba,andP.G.Ifju,\Enab lingAutonomousFlight,"in Proc.FloridaConferenceontheRecentAdvancesin Robotics ,DaniaBeach,August2003,vol.16,sec.TA3,pp.1{3. [3]AeroVironment,\AUV:Pointer,"WorldWideWeb, http://www. aerovironment.com/area-aircraft/prod-serv/pointer.h tml ,lastaccessedonNovember15,2004. [4]J.M.GrasmeyerandM.T.Keennon,\DevelopmentoftheBla ckWidow MicroAirVehicle,"in AIAA-2001-127 ,January2001. [5]P.G.Ifju,S.Ettinger,D.A.Jenkins,Y.Lian,W.Shyy,an dM.R.Waszak, \Flexible-wing-basedMicroAirVehicles,"in 40thAIAAAerospaceSciences Meeting ,Reno,Nevada,January2002. [6]P.G.Ifju,S.Ettinger,D.A.Jenkins,andL.M.ez,\Compo siteMaterialsfor MicroAirVehicles,"in SAMPEJournal ,July/August2001,vol.46/2,pp. 1926{1937. [7]D.A.Jenkins,P.G.Ifju,M.Abdulrahim,andS.Olipra,\A ssessmentof ControllabilityofMicroAirVehicles,"in Proc.16thIntlConf.UnmannedAir VehicleSystems ,April2001,vol.17,pp.617{640. [8]S.M.Ettinger,\DesignandImplementationofAutonomou sVision-guided MicroAirVehicles,"M.S.thesis,UniversityofFlorida,Ma y2001. [9]S.Todorovic,\StatisticalModelingandSegmentationo fSky/GroundImages," M.S.thesis,UniversityofFlorida,December2002. [10]S.Todorovic,M.C.Nechyba,andP.G.Ifju,\Sky/Ground Modelingfor AutonomousMAVs,"in Proc.IEEEInt.ConferenceonRoboticsand Automation ,NewOrleans,September2003,vol.1,pp.1422{1427. 32

PAGE 40

33 [11]T.D.Morton, EmbeddedMicrocontrollers ,Prentice-Hall,Columbus,2001. [12]J.Plew,J.W.Grzywna,M.C.Nechyba,andP.G.Ifju,\Rec entProgressin theDevelopmentofOn-BoardElectronicsforMicroAirVehic les,"in Proc. FloridaConferenceontheRecentAdvancesinRobotics ,Orlando,August 2004,vol.17,sec.FP3,pp.1{6. [13]J.Plew,\DevelopmentofFlightAvionicsSystemforAut onomousMAV Control,"M.S.thesis,UniversityofFlorida,December200 4. [14]J.W.Grzywna,A.Jain,J.Plew,andM.C.Nechyba,\Rapid Developmentof Vision-BasedControlforMAVsthroughaVirtualFlightTest bed,"submitted toIEEEInt.Conf.onRoboticsandAutomation,April2005. [15]M.Quigley,M.A.Goodrich,andR.W.Beard,\Semi-Auton omousHumanUAVInterfacesforFixed-WingedMini-UAVs,"tobepresente datIEEEInt. Conf.onIntelligentRobotsandSystems,August2005. [16]S.M.Ettinger,M.C.Nechyba,P.G.Ifju,andM.Waszak,\ Vision-guided FlightStabilityandControlforMicroAirVehicles,"in ProceedingsIEEEInt. ConferenceonIntelligentRobotsandSystems ,Lausane,October2002,vol.3, pp.2134{2140. [17]B.Nadel,\WirelessNetworking101,"WorldWideWeb, http://reviews. cnet.com/4520-3243_7-5021297.html ,lastaccessedonNovember15,2004. [18]D.A.ForsythandJ.Ponce, ComputerVision:AModernApproach ,PrenticeHall,UpperSaddleRiver,2003. [19]A.Kurdila,M.C.Nechyba,R.Lind,P.G.Ifju,W.Dahmen, R.DeVore, andR.Sharpley,\Vision-basedControlofMicroAirVehicle s:Progressand ProblemsinEstimation,"in PresentedatIEEEInt.ConferenceonDecision andControl ,Atlantis,2004. [20]L.Armesto,S.Chroust,M.Vincze,andJ.Tornero,\Mult i-rateFusionwith VisionandInertialSensors,"in Proc.IEEEInt.ConferenceonRoboticsand Automation ,April2004,vol.1,pp.193{199. [21]R.C.DorfandR.H.Bishop, ModernControlSystems ,Addison-Wesley, MenloPark,1998.

PAGE 41

34 [22]J.W.Grzywna,\AFlightTestbedwithVirtualEnvironme ntCapabilities forDevelopingAutonomousMicroAirVehicles,"M.S.thesis ,Universityof Florida,December2004.

PAGE 42

BIOGRAPHICALSKETCH UrielRodriguezwasborninPonce,PuertoRico,onJune6,197 9.Hisfamily movedtoCoralSprings,FL,in1988.Hereceivedhishighscho oldiplomafrom MarjoryStonemanDouglasHighSchoolinParkland,FL.Hethe nattendedthe UniversityofFloridaandreceivedhisbachelor'sdegreein computerengineering inDecember2001.Whileworkingonhisundergraduatework,U rielwasactivein theStudentIEEEBranchandwasamemberoftheIEEESoutheast ConStudent HardwareTeamwhichwon1stPlacewithitsPongrobot.Sincet henUrielhas workedintheMachineIntelligenceLabunderDr.AntonioArr oyo,Dr.Michael Nechyba,andDr.EricSchwartz. 35


Permanent Link: http://ufdc.ufl.edu/UFE0008480/00001

Material Information

Title: Miniaturization of Ground Station for Unmanned Air Vehicles
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0008480:00001

Permanent Link: http://ufdc.ufl.edu/UFE0008480/00001

Material Information

Title: Miniaturization of Ground Station for Unmanned Air Vehicles
Physical Description: Mixed Material
Copyright Date: 2008

Record Information

Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
System ID: UFE0008480:00001


This item has the following downloads:


Full Text











MINIATURIZATION OF GROUND STATION FOR UNMANNED AIR
VEHICLES
















By

URIEL RODRIGUEZ


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF SCIENCE

UNIVERSITY OF FLORIDA


2004















TABLE OF CONTENTS

page

LIST OF FIGURES ................................ iv

A B STR A CT . . . . . . . . . vi

1 INTRODUCTION .............................. 1

1.1 M otivation . . . . . . . 1
1.2 Previous W ork . . . . . . . 2
1.3 C('! !!. iges ... . . . .. .... .. ... 3
1.4 O verview . . . . . . . . 4

2 GROUND STATION ............................. 5

2.1 Previous W ork . . . . . . . 5
2.2 Handheld Ground Station . . . . ... 7
2.2.1 Motivation for Miniaturization . . ... 7
2.2.2 Selection and Specifications of the Handheld Ground Station 7
2.2.3 Limitations . . . . . . ... 8

3 HORIZON ESTIMATION . . . . . . 10

3.1 Previous W ork . . . . . . . 10
3.2 Our Approach . . . . . . . 11
3.2.1 Overview . . . . . . . 11
3.2.2 Algorithm Details . . . . . 12
3.2.3 Algorithm Performance . . . . . 16

4 TESTBED AND EXPERIMENTAL RESULTS ... . 20

4.1 Testbed . . . . . . . . 20
4.1.1 Testbed Setup . . . . . . 20
4.1.2 Controller . . . . . . 22
4.2 R results . . . . . . . . 23
4.2.1 Virtual Testbed . . . . . . 23
4.2.2 Flight Testing . . . . . . 27















5 FUTURE WORK ....


5.1 Ground Station

5.2 Vision Algorithm


REFERENCES . .


BIOGRAPHICAL SKETCH.


...........................


...........................

...........................


...........................


...........................















LIST OF FIGURES


Figure page

2.1 The folding-wing "Pocket MAV" can be stored in the container shown. 6

2.2 Size comparison of 12" Powerbook to Zaurus. . ...... 8

3.1 Visual representation of algorithm. ........ . 11

3.2 Example of Ettinger's initial attempt failing. .... . . 12

3.3 The result of our algorithm on a sample image is shown above. (a)
The estimated horizon is shown in red. (b) Blue pixels were classi-
fied as sky while green pixels were classified as ground. .. . 13

3.4 The output of the bootstrap process is shown above. (a) Bootstrap
image from a real flight test. (b) Distribution of ground and sky
pixels for the given bootstrap image. ....... . . 13

3.5 Output of our algorithm for a bank angle greater than 45 degrees.
(a) Estimated horizon shown in red. (b) Classified pixels and es-
timated horizon . . . . . . . 15

3.6 Correct estimation of horizon. . . . . . 17

3.7 The horizon was properly estimated in both images even though both
im ages were noisy . . . . . . 18

3.8 The results of skipping pixels are shown above. (a) Classifying every
other 5 lines. (b) Classifying every other 20 lines. . ... 19

4.1 Testbed setup . . . . . . . 21

4.2 Linear estimator. 3 degrees of error at worst between -45 and 45 de-
grees . . . . . . . . 22

4.3 Various images from the flight simulator software. . . 24

4.4 Controller output . . . . . . . 25










4.5 Image sequence from a virtual environment test flight. . ... 25

4.6 MAV flown in tests . . . . . . . 26

4.7 Sample images taken from the flight test. . . . . 27

5.1 The disparity between CCD and C\ I OS image quality is shown above.
(a) Image taken with C'\!OS camera. (b) Image taken with CCD
cam era . . . . . . . . 31















Abstract of Thesis Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Master of Science

MINIATURIZATION OF GROUND STATION FOR UNMANNED AIR
VEHICLES

By

Uriel Rodriguez

December 2004

C'!i i': Amauri A. Arroyo
M, iPr Department: Electrical and Computer Engineering

In this thesis, we seek to miniaturize the ground station responsible for the

image processing that is used to autonomously stabilize unmanned air vehicles

(UAVs). The image processing performed on the ground station consists of finding

the horizon in the images transmitted from the UAV's forward-facing camera. In

conjunction with a proportional derivative controller, the estimated horizon is then

used to stabilize the UAV. Thus, the focus of our research is to derive a horizon

estimating algorithm that can be executed in real-time on a smaller ground station.

To achieve our goals, we propose to use a ground station based on a handheld

computer, also known as personal digital assistant (PDA). The PDA-based ground

station is considerably smaller than the previous ground station which consisted of

a notebook computer. However, the lack of a floating-point unit (FPU) hinders the

development of a real-time horizon estimation algorithm.










To address the lack of an FPU, we propose the use of a non-floating point

intensive horizon estimation algorithm. The overall approach of the algorithm is to

find the transition between sky and ground in each vertical line of the image. The

classification of sky and ground is achieved by comparing the pixel in question to

the color models for both the sky and the ground. The pixel is then classified as

sky or ground depending on its nearest color model. Once the positions of all the

transitions are known, linear regression is performed on the transition points to

estimate the horizon in the image.

Our initial results support the outlined approach. The proposed algorithm

successfully estimates the horizon in real-time environments.















CHAPTER 1
INTRODUCTION


1.1 Motivation

A major focus of the United States Air Force is the development of unmanned

air vehicles (UAVs) that can be deploy, ,1 in strategic operations as well as tactical

scenarios including the ability for soldiers in the field to deploy micro air vehicles

(\ !AVs). A MAV is an UAV with a small wing span that flies at low altitudes and

slow airspeeds. A MAV consists of a wingspan and fuselage that ranges from 30

inches down to six inches and operates at speeds less than 25 miles per hour [1].

The MAVs offer great potential in both the military and civilian sectors.

Equipped with small cameras and transmitters, MAVs can be used to survey and

monitor areas that are too far or too dangerous to send human scouts. Possible

civilian applications include monitoring radiation spills, forest fires, and volcanic

activity. In the military, MAVs are intended to reduce the risk of personnel by

assisting ground troops in search and rescue missions, tracking remote moving-

targets, and assessing immediate bomb damage.

However, there are certain limitations that have restricted the wide deployment

of MAVs. A MAV requires highly skilled pilots to fly such small planes. The small

size of the aircraft tends to make the MAVs more susceptible to wind gusts,

and thus, harder to stabilize and control. This problem can be remedied by the

development of an autonomous MAV that is self-stabalizable [2]. However, the










ground station required to deploy an autonomous MAV is bulky and would disallow

the use of a self-stabilizing MAV in certain situations. The development of a

smaller ground station would allow for the deployment of autonomous MAVs in

wider variety of scenarios.

1.2 Previous Work

There have been numerous successful UAVs implemented by various companies

including Lockheed Martin, Northrop Grumman, and AeroVironment. A notable

UAV is the Pointer and is developed by AeroVironment [3]. The Pointer was

designed as a tactical reconnaissance vehicle for military and law enforcement

applications. An onboard camera reJ-,v live video images to a ground station that

the pilot can use to control the UAV remotely. Although the Pointer is currently

being used by the military, its large size keeps it from being deploy, ,1 in certain

situations. For situations requiring a smaller UAV, AeroVironment developed an

autonomous MAV named the Black Widow [4]. Like the Pointer, the Black Widow

is capable of autonomous flight and is equipped with a camera whose video is

transmitted to the ground.

At the University of Florida, the MAV Lab has become very proficient in the

research and development of such small aircraft [5-7]. The MAV Lab continues to

develop planes with improved flight characteristics, p loadd capacity, and structural

integrity. The Lab has developed planes that have won the International MAV

Competition for the past six consecutive years. The competition judges entries

based on the size and the operational range of the aircraft. The planes of the MAV

Lab were originally controlled by off the shelf remote control airplane equipment,

and therefore, controlled completely by a human pilot. Since the operational range










of the plane was limited by the ability of the pilot to effectively follow the small

plane in the air, the planes were fitted with forward-looking cameras and video

transmitters that would allow the pilot to fly the planes well beyond visual range.

The computer vision research at the University of Florida took advantage

of the fact that the MAVs were sending a video signal to the ground. The initial

vision system developed analyzed the images obtained from the forward-looking

camera in order to find the horizon [8, 9, 10]. In conjunction with a proportional

integral derivative (PID) controller, the horizon was used to determine the neces-

sary adjustments needed to keep the plane level with respect to the ground. Once

implemented in real-time, the vision-based flight stability system could keep a MAV

in the air without any input from a human pilot. The advantage of an autonomous

MAV is that it allows unexperienced pilots to fly them since the MAVs are capable

of self-stabilization.

1.3 Challenges

Initially, the image processing, required to estimate the horizon, was performed

on the ground using a desktop computer. As technology improved, the vision

algorithm was ported to a ground-based notebook computer. However, if ground

troops are to deploy MAVs on the battlefield, then it is essential that the ground

station be as compact as possible.

A number of formidable challenges have been encountered when miniaturizing

the ground station. These challenges result from features that are unavailable

on current handheld computers. These missing features include the lack a fast

input/output (I/O) port and the lack of a floating-point unit (FPU).










In previous ground stations, the FireWire port (IEEE 1394) was used in

conjunction with a frame grabber to import the incoming video stream to the

ground station. However, handheld computers are currently lacking a FireWire port

or an equivalently fast I/O port like the second generation Universal Serial Bus

(USB2). The proposed approach is to use the compact flash card slot to import the

video stream via a wireless Internet connection.

Since floating-point instructions take about an order of magnitude longer to

execute on a processor without an FPU [11], the lack of an FPU will limit the

ability of the vision system. A non-floating point intensive horizon estimation

algorithm is proposed.

1.4 Overview

In this thesis, we first introduce the motivation and challenges for miniaturiz-

ing the ground station. C'!i Ipter 2 discusses the need for a smaller ground station

and implications of migrating the ground station to a handheld computer. C'!I Ipter

3 describes the horizon detection algorithm designed to overcome the shortcomings

of the handheld computer. C'!, Ipter 4 describes the tested setup and illustrates the

experimental results of autonomous flights in a simulator as well as in real flights.

Finally, C'!i Ipter 5 offers some ideas for future work.















CHAPTER 2
GROUND STATION


In this Ch'! pter, we first describe the evolution of the ground station and the

motivation to shrink the ground station even further. We then propose the use of a

handheld computer as the next ground station. Finally, we examine the limitations

encountered when using a handheld computer as the ground station.

2.1 Previous Work

Our ultimate goal is to perform the horizon estimation, used to stabilize the

aircraft, on-board the MAV [12, 13]. On-board image processing would eliminate

the transmission noise problems and frame dropouts caused by transmitting

the video to the ground for processing. However, because of the computational

demands of the horizon estimation and the p .,load limitations of the MAV, the

calculations presently need to be performed on the ground.

Both the Pointer and the Black Widow use a ground station that pilots use to

control the UAVs. The Pointer ground station is the size of a briefcase 7" x 11"

x 16" [3]. The Black Widow went through several iterations of its ground station

[4]. It started as a collection of off-the-shelf components that had to be assembled

on the field and evolved into the size of a 15-lb briefcase. Eventually, the Black

Widow's ground station shrunk to the size a compact Pelican case.



























Figure 2.1: The folding-wing "Pocket MAV" can be stored in the container shown.

At the University of Florida, the first ground station consisted of a 900MHz

desktop computer with a frame grabber card running the Mandrake Linux operat-

ing system [8]. As computer technology improved over the years, the ground station

was further reduced from a desktop computer to a 12" Apple Powerbook notebook

computer running at 1GHz [14]. Switching to an Apple Powerbook removed the

need for a frame grabber card since the notebook is equipped with a FireWire

400 (IEEE1394a) port and a QuickTime API that provides an easy interface to

capture images from the FireWire port. Another advantage of this Apple computer

is its operating system OS X. OS X is UNIX-based operating system, and thus,

allows for easy migration of existing code from a Linux environment to the OS X

platform.










2.2 Handheld Ground Station

2.2.1 Motivation for Miniaturization

In 2003, the Mechanical and Aerospace Engineering Department at the

University of Florida received a grant from the U.S. Army that would require the

development of a plane with a small storage footprint. This grant was for the

development of a plane and the ground station that could fit in the pocket of a

soldier's battle dress uniform (BDU). The BDU's pocket dimensions are 8" x 8" x

2". The MAV Lab produced a folding-wing airplane that when folded, can fit in

a container that is 8" x 4" x 2". The remaining 8" x 4" x 2" of unoccupied space

in the BDU is too small to accommodate the 12" Powerbook. Therefore, a smaller

ground station needs to be developed. Both the MAV and the container are shown

in Figure 2.1.

2.2.2 Selection and Specifications of the Handheld Ground Station

The Sharp Zaurus SL-5600 is the handheld computer, also known as a personal

digital assistant (PDA), selected to become the next ground station. The Zaurus

specifications are shown in Table 2.1. The main advantage of the Zaurus is its

Embedded Linux Operating System. A Linux-based operating system is an

advantage because once again, any previously designed code can be cross-compiled

for this device. It also provides a familiar set of development tools, including, the

GNU Development Tools.

A more obvious advantage of the handheld is its size. Measuring 0.9" x

5.4" x 2.9", the Zaurus is more than 12 times smaller in volume than the 12"

Powerbook (1.35" x 12.7" x 10.2"). At 7.1 oz, the Zaurus weighs 13 times less than





























Figure 2.2: Size comparison of 12" Powerbook to Zaurus.

the Powerbook (5.9 lbs). Figure 2.2 illustrates the physical difference between the

Zaurus and the Powerbook.

2.2.3 Limitations

The Zaurus adds two i i" r limitations. Even though the Zaurus utilizes

Intel's XScale processor running at 400MHz, it is lacking a Floating-Point Unit

(FPU). Since floating-point operations must be performed in software on the

Table 2.1: Specifications of the Sharp Zaurus SL-5600.

CPU Intel XScale (PXA255, 400MHz)
Platform Linux 2.4 (Embedix)
Di'p1lv 3.5" Reflective TFT Color Dipliv with 240x320 Resolution
Memory 32MB SDRAM / 6 I \i H Flash ROM
I/O Compact Flash Slot, SD Card Slot, Serial Port










Zaurus, floating-point operations place a huge burden on the processor, taking up

to an order of magnitude longer to execute than integer operations [11]. Therefore,

the algorithm performed on the Zaurus should be designed to avoid using floating-

point calculations.

Another limitation of the Zaurus SL-5600 is the lack of FireWire or USB ports.

This eliminates the possibility of using a FireWire or USB frame grabber to import

the video feed from the plane. An 802. lb Wireless Compact Flash Card on the

Zaurus and a host computer with a FireWire port and network access are used to

import the images in to the Zaurus.

Brigham Young University (BYU) has successfully used PDAs as an interface

to UAVs [15]. BYU employs a similar method of using a host computer to commu-

nicate to the PDA via an 802.11b link. It is important to note that BYU has only

used the PDA as an interface to the MAV while still using the host computer to

handle the data processing and communication to the UAV.

In the next C!i Ipter, we will discuss the previous attempts to detect the

horizon and describe an algorithm that will successfully estimate the horizon on a

PDA in real-time.















CHAPTER 3
HORIZON ESTIMATION


In this C'! pter, we discuss the previous work done to estimate the horizon.

We also propose a new approach that will allow us to estimate the horizon on a

handheld computer. Finally, the performance of the new approach is discussed.

3.1 Previous Work

In the initial attempt to estimate the horizon, the main assumption was that

the sky would be represented by high-intensity (light) image region while the

ground would be represented by the low-intensity (dark) image region [8]. Locating

the boundary between the sky and the ground would result in the horizon. The

basic idea behind the algorithm was to fit a step function to the intensity values

of each vertical line in the image as shown in Figure 3.1. Once the position of the

best-fit transitions for each vertical line in the image are known, a linear regression

was performed on the set of (x,y) positions. The resulting line would estimate

the horizon. Although this approach worked when flying over a uniform forest on

a sunny d4w, it resulted in large errors in the horizon estimates when flying over

ground objects of high intensity. An example of this error is shown in Figure 3.2.

Since only marginal results were obtained from the first algorithm, a different

approach was taken to correctly estimate the horizon. Instead of combining the

color information into intensity, and thus, losing two-thirds of the information avail-

able in the process, both sky and ground were modeled as a statistical distribution
















I'l"


Figure 3.1: Visual representation of algorithm.

in color space [16]. The task was then to find the set of points within the image
that would have the highest likelihood of fitting the given distributions. This is
accomplished by performing a search through the potential set of all lines in the
image space and finding the line that best fit the distributions.
Although the color model-based horizon detection algorithm was demonstrated
to work at 30Hz in real-time with over 99.9' correct horizon identification, it will
not run anywhere close to 30Hz in real-time on the Zaurus because of its reliance
on floating-point calculations [16].

3.2 Our Approach

3.2.1 Overview

In order to stabilize the MAV with the Zaurus, a robust, yet non-floating-point
intensive horizon detection algorithm must be developed. The new algorithm


























Figure 3.2: Example of Ettinger's initial attempt failing.

borrows from both of the algorithms discussed above to create a lightweight, yet

effective horizon detection system.

The overall approach of the algorithm is to find the transition between sky

and ground in each vertical line. Pixels are classified according to its nearest sky

or ground color model. Once all the (x,y) positions of the transitions for each

vertical line are known, a linear regression is performed to estimate the horizon as

illustrated in Figure 3.3.

3.2.2 Algorithm Details

The first step in the estimation of the horizon is to be able to classify each

pixel in the image as either sky or ground. A color-based model is built of both

the sky and the ground. Not surprisingly, it turns out that the blue and green

channels of an image containing a horizon produce a distinct distribution of the

sky and the ground pixels. A bootstrap procedure is performed before each aircraft




























(a) (b)


Figure 3.3: The result of our algorithm on a sample image is shown above. (a) The
estimated horizon is shown in red. (b) Blue pixels were classified as sky while green
pixels were classified as ground.


16!

14C

^. 1203





06c

4C


0 20 40 60 80 100 120C 140 160
Blus Intenralty
(b)

Figure 3.4: The output of the bootstrap process is shown above. (a) Bootstrap im-
age from a real flight test. (b) Distribution of ground and sky pixels for the given
bootstrap image.










launch to build the color model. This is accomplished by pointing the plane's

camera towards the horizon and assuming that the top one-third of the image is

sky and the bottom third of the image is ground. Figure 3.4 shows the color model

distribution results of the bootstrap sequence.

Once the color models are acquired, the image is traversed in the vertical

direction. As each column is traversed, each pixel is classified as sky or ground

depending on its nearest color model. The starting and ending locations of the

largest consecutive set of sky and ground pixels in each column are also recorded.

Next, the transition between sky and ground is determined by the sky pixel from

the largest consecutive set of sky pixels that is closest to the largest consecutive

set of ground pixels. This is based on the assumption that ground pixels are more

likely to be classified as sky pixels than sky pixels being classified as ground pixels.

This algorithm does not assume that every vertical line will include a transi-

tion from sky to ground, as would be the case in an image with a roll angle greater

than 45 degrees. Therefore, the algorithm must determine when a transition has

not occurred and discard the transition point generated by that line. A good tran-

sition must meet the following two requirements: (1) The end of the largest set of

sky pixels that is closest to the edge of the image must be close enough to the edge

of the image, and (2) the other edge which is nearest to the middle of the image

must be far enough from the edge.

It should be evident that there will be cases in which only a few good transi-

tions will be found by the algorithm. This again is the case in an image with a roll

angle greater than 45 degrees. One way to approach this problem is to calculate an

initial horizon estimate, then if the roll angle is greater than 45 degrees, repeat the







15













(a) (b)

Figure 3.5: Output of our algorithm for a bank angle greater than 45 degrees. (a)
Estimated horizon shown in red. (b) Classified pixels and estimated horizon.


process using the horizontal lines in order to increase the accuracy of the horizon.

However, this approach would require the linear regression to be performed twice,

once for the vertical lines and once for the horizon lines. The linear regression

calculation are floating-point intensive, and therefore, would hurt the overall per-

formance. Moreover, it might be possible not to be able to estimate a horizon from

the vertical lines if the roll angle is close to 90 degrees. The alternate approach

used to eliminate the extra floating-point calculations is to keep track of the num-

ber of good transitions. Based on the number of good transitions found and the

size of the image, it is possible to determine if the process needs to be performed on

the horizon lines. The result of a horizontal sweep is shown in Figure 3.5.

Once the pixels have been classified and the minimum number of good

transition points have been identified, linear regression is performed on the good

transitions to estimate the horizon. Linear regression outputs the slope, m, and

the y-intercept, b, of the line that corresponds to the estimated horizon and is









calculated by using the following formulas:

nE(xy) ExEy
S E(x) (E )(31)
LLE( 2) ExEx_
n (x2) (x (3.2)


where x and y represent the (x,y) coordinates of good transitions identified in the

image. The roll angle (in degrees), Q, can be calculated by the inverse tangent of

the slope of the horizon line:

Satan(m) x 180/7. (3.3)

This algorithm also allows a way to easily determine the orientation of

the aircraft at all times. While performing vertical scans, if the largest set of

consecutive sky pixels is above the largest set of consecutive ground pixels, then

the plane is flying normal. If the largest set of consecutive sky pixels is below

the largest set of consecutive ground pixels, then the plane is flying upside down.

Similarly, it is possible to determine the location of the sky relative to the ground

while performing horizontal scans when the plane is in a steep roll angle.

3.2.3 Algorithm Performance

The performance of our algorithm is more robust than the first algorithm

discussed in the beginning of this C'!i pter. As previously shown in Figure 3.2,

the intensity-based algorithm failed to correctly estimate the horizon where high-

intensity objects were present on the ground. Figure 3.6 shows the correctly

estimated horizon achieved by using our algorithm on the same image. Our




























Figure 3.6: Correct estimation of horizon.


algorithm can also correctly estimate the horizon in noisy images as shown in

Figure 3.7.

Our algorithm also executes much faster in a CPU without an FPU than the

previous algorithms since the new approach uses significantly less floating-point

operations. Performance can be further enhanced by skipping lines. If every other

line is skipped, the vertical scan time would be decreased by one-half. If every third

vertical line is analyzed, then the vertical scan time would be decreased by a third,

and so forth. The difference between the output produced by analyzing each line

and the output produced by skipping lines is negligible. The algorithm results of

skipping pixels is shown in Figure 3.8. The amount of pixels that are skippable

depends on the size of the image. For a larger image, more lines can be skipped.

In the next C'! ipter, we will discuss the testbed setup used to test the al-

gorithm in a real-time environment. This setup will include flights in a virtual






















Figure 3.7: The horizon was properly estimated in both images even though both
images were noisy.

environment as well as real test flights. Finally, the results of both the virtual and

real test flights will be shown.


































(a) (b)
Figure 3.8: The results of skipping pixels are shown above. (a) Classifying every
other 5 lines. (b) Classifying every other 20 lines.















CHAPTER 4
TESTBED AND EXPERIMENTAL RESULTS


In this C'! ipter, we discuss the testbed setup used to test the vision algorithm

in a real-time environment. We show how a virtual environment is used to verify

the algorithm before a real test flight takes place. Finally, the results of the flight

tests are discussed.

4.1 Testbed

4.1.1 Testbed Setup

A pictorial diagram of the testbed setup is shown in Figure 4.1. The video

signal is transmitted from the airborne MAV to the ground via a 2.4GHz RF

transmitter. The signal is then fed into a laptop via a FireWire frame buffer that

down-samples the image to 80x60 resolution. The laptop then transmits the down-

sampled image to the Zaurus via an 802.11b link for image processing. After the

horizon is estimated on the Zaurus, the desired servo positions are transmitted

via an RS-232 link to an 8-bit micro-controller that converts them to a servo

train-pulse that is then fed to the RC controller. When the trainer switch on the

controller is engaged, the controller transmits the train-pulse generated from the

desired servo positions. When the trainer switch is not engaged, then a human pilot

has control of the plane. This is beneficial for flight testing in case something goes

wrong, and the human has to retake control of the plane to avoid a potential crash.






















.1._l.--- Signal Generator |




Figure 4.1: Testbed setup.

The image needs to be down-sampled to 80x60 resolution or smaller, not

because of the performance of the algorithm, but because of the bottleneck caused

by the bandwidth of the 802.11b connection. An 8-bit depth, uncompressed

RGB image arriving at 30Hz would need a bandwidth of 3., i\!l)ps (8 bits x 3

color channels x 80 width x 60 height x 30 times per second). Although the

802.11b protocol has a raw rate of to up 11Mbps, its actual throughput is about

half of that [17]. Also, as signal strength and quality decreases, the throughput

diminishes. If the image size is increased to 120x80, then the required bandwidth

would be 6.7Mbps which is more than the 5.5Mbps actual throughput bandwidth

of 802.11b. However, there is a positive side-effect to down-sampling the image.

The down-sampling acts as a mean filter [18], so there are less outliers in the color

distribution.































Figure 4.2: Linear estimator. 3 degrees of error at worst between -45 and 45 de-
grees.

4.1.2 Controller

In the past, complicated controllers have been used to stabilize the plane

[19, 20]; however, a simpler approach is taken in the experiments performed. A

simple proportional derivative (PD) controller is used to correct the roll of the

plane [21]. Equation 4.1 is used to calculate the desired servo position, 6d, where

Kp and Kd are the proportional and derivative gains, respectively, and 6," is the

neutral servo position. The roll angle, 4, calculation can be simplified from a

floating-point inverse tangent (4.2) to an integer multiplication (4.3) via a linear










estimator.


d = K- Kd, (4.1)

= arctan(m) x 180/7, (4.2)

x 48. (4.3)


The linear estimator for the roll angle works best for angles between -45 and

45 degrees where the error is less 4 degrees as shown in Figure 4.2. A non-linear

estimator can easily be built to produce small errors for all possible roll angles.

The fact that the linear estimator causes a large error at angles greater than 45

degrees is not a problem since the effect that the error causes is a larger correction

rate.

4.2 Results

4.2.1 Virtual Testbed

A virtual environment is used to test the hardware, software, and algorithm

before an actual test flight [22]. The virtual tested is based on an off-the-shelf

remote control airplane simulation package. This setup uses the same interface to

the ground station that the MAV uses, and thus, allows for seamless interchanging

between the two. Instead of transmitting the plane's video to the ground station,

the simulator's video output is fed into the ground station. Similarly, the RC

controller's output becomes the input to the airplane simulator. The simulation

package offers a diverse set of scenery that allows for to the testing of hardware,

horizon estimation, and the controller without the need to fly a MAV. Figure 4.3












































Figure 4.3: Various images from the flight simulator software.













250

200-

150-

100 -









-100 -

-150- -R





Figure 4.4: Controller output.


Figure 4.5: Image sequence from a virtual environment test flight.


li





























Figure 4.6: MAV flown in tests.


illustrates some of the environments available in the simulation package as well as

the results from the horizon estimator.

Although the initial results as shown in Figure 4.3 are good, a lag has been

encountered that rendered the system unusable in a real-time environment. The lag

is introduced by the time required to send the image from the host computer to the

Zaurus via 802.11b. By the time that the first image is transferred to the Zaurus,

the next image is ready to be transferred, and thus, leaving virtually no time for the

image processing. This problem is remedied by only transferring every other frame

to the Zaurus. This technique diminishes the effect of the lag and allows the entire

process to execute in real-time at 15Hz.

The controller performs well while running in real-time at 15Hz. The servo

command sent to the simulation as produced by the controller to stabilize the plane




























Figure 4.7: Sample images taken from the flight test.

is show in Figure 4.4. Figure 4.5 shows a sequence of images where the plane is

stabilizing the horizon.

4.2.2 Flight Testing

After finalizing the system parameters in the virtual environment, the ground

station is ready for a real test flight. The UAV flown for this experiment is shown

in Figure 4.6. On its maiden flight, the ground station produced excellent results.

The PDA-based ground station correctly estimated the horizon and successfully

stabilized the plane in real-time. The controller gains used in the virtual tested

worked on the real plane without requiring any additional adjustments. The images

shown in Figure 4.7 are the result of the horizon estimator on images taken from

the test flight.

An unforeseen advantage was encountered while flight testing. In the past, the

pilot has been constrained to flying the plane just a few feet away from the ground







28

station because the pilot's RC controller was used to send commands to the plane.

Now, the pilot is free to roam as he is flying the plane because only the PDA is

attached to his RC controller. The other parts of the ground station which include

the antenna, receiver, frame grabber, and laptop can be located ., i-v from the pilot

since the data is transmitted to the PDA wirelessly.















CHAPTER 5
FUTURE WORK


The systems developed in this work have only taken initial steps toward a

small, stand-alone ground station. This C'!i pter will discuss improvements to both

the ground station and the vision algorithm that will allow us to meet those goals.

5.1 Ground Station

As technology improves, the potential and importance of the PDA as a stand-

alone ground station for a UAV will increase. There are already better PDAs in

the market sporting a 62, I\1II. CPU and 6 0!\!1 of SDRAM. That is already a 6 !'.

increase in processing power and double the amount of RAM. A new transflective

TFT VGA display with 640x480 resolution is a huge improvement over the old

320x240 resolution reflective TFT di- ppl which had poor visibility in direct

sunlight. The new PDAs also carry a 2D and 3D video accelerator with 16MB of

video memory. The video accelerator will greatly decrease the amount of CPU

processing power required to repaint the screen when displaying the incoming

images to the PDA display.

Although some PDAs have the capability to act as a USB host, and thus, allow

the possible use of a USB frame grabber that would eliminate the need for laptop

computer, the limitation lies in the inability to easily customize the operating

system. Complete embedded Linux distributions, such as OpenZaurus, are not

yet easy to configure and customize, but will be available in the near future. This










ability would not only be beneficial to the overall performance of the system by

trimming needless features from the kernel and other unnecessary applications, but

it would also be needed to compile Video4Linux and other features into the kernel

which are required by most frame grabber modules.

Ideally, FireWire support would be added to PDAs. Unfortunately, there is not

much demand for such a feature. The CPUs would probably need to reach speeds

of at least 1GHz before such a fast data bus can be fully utilized. After all, there's

no need to transfer information so quickly into a system, if the processor cannot

keep up with it. A Floating-Point Unit would also be a welcomed edition to the

PDA, but again, there is not much demand for it in the current market. Hopefully,

as PDAs become more and more powerful, they will start replacing laptops, and

thus, in the process be retrofitted with some of the features that they currently

lack.

5.2 Vision Algorithm

Although the vision algorithm was successful in estimating the horizon, steps

can be taken to improve its results. The performance of the vision algorithm is

directly correlated to its ability to classify sky and ground pixels. Better quality

images would lead to better color models which in turn would increase the per-

centage of correctly identified sky and ground pixels. A CCD (charged coupled

device) camera on the aircraft would produce more vivid, lower-noise images than

the current C`\ I OS (complementary metal oxide semiconductor) cameras used for

the flight tests. The light sensitivity of C\ !OS cameras is much lower than CCD

cameras. Figure 5.1 shows the dramatic difference between an image captured with

a C'\ OS camera and an image captured with a CCD camera on the same dw. A






















(a) (b)
Figure 5.1: The disparity between CCD and C'\!OS image quality is shown above.
(a) Image taken with C \!OS camera. (b) Image taken with CCD camera.

dynamic color model could also boost classification results as lighting conditions

change throughout the d4w.

More traditional image processing technics like erosion and dilation can reduce

unwanted noise or misclassifications in an image. Also, a blob-finding technic to

find and eliminate small sets of connected pixels can be used similarly to erosion

and dilation to improve the overall quality of the image after classification.

The controller could also be greatly improved. A model-based controller using

state feedback from the vision system would be a tremendous improvement over

the current proportional-derivative controller approach. The controller could also

implement joystick control from the PDA to use to directly manipulate the plane.

Along with a compact flash modem, this would eliminate the need for the 8-bit

micro-controller and the RC controller.














REFERENCES


[1] J. M. McM. and Col. M. S. Francis, \i, ro Air Vehicles Toward a New
Dimension in Flight," World Wide Web, http://www. darpa.mil/tto/mav/
mav_auvsi .html, last accessed on November 15, 2004.

[2] J. W. Grzywna, J. Plew, M. C. Nechyba, and P. G. Ifju, "Enabling Au-
tonomous Flight," in Proc. Florida Conference on the Recent Advances in
Robotics, Dania Beach, August 2003, vol. 16, sec. TA3, pp. 1-3.

[3] AeroVironment, "AUV: Pointer," World Wide Web, http: //www.
aerovironment. com/area-aircraft/prod-serv/pointer .html, last ac-
cessed on November 15, 2004.

[4] J. M. Grasmeyer and M. T. Keennon, "Development of the Black Widow
Micro Air Vehicle," in AIAA-2001-127, January 2001.

[5] P. G. Ifju, S. Ettinger, D. A. Jenkins, Y. Lian, W. Shyy, and M. R. Waszak,
"Flexible-wing-based Micro Air Vehicles," in 40th AIAA Aerospace Sciences
Meeting, Reno, Nevada, January 2002.

[6] P. G. Ifju, S. Ettinger, D. A. Jenkins, and L. M.ez, "Composite Materials for
Micro Air Vehicles," in SAMPE Journal, July/August 2001, vol. 46/2, pp.
1926-1937.

[7] D. A. Jenkins, P. G. Ifju, M. Abdulrahim, and S. Olipra, "Assessment of
Controllability of Micro Air Vehicles," in Proc. 16th Intl Conf. Unmanned Air
Vehicle S.-/1' '- April 2001, vol. 17, pp. 617-640.

[8] S. M. Ettinger, "Design and Implementation of Autonomous Vision-guided
Micro Air Vehicles," M.S. thesis, University of Florida, M i,- 2001.

[9] S. Todorovic, "Statistical Modeling and Segmentation of Sky/Ground Images,"
M.S. thesis, University of Florida, December 2002.

[10] S. Todorovic, M. C. N. i. vba, and P. G. Ifju, "Sky/Ground Modeling for
Autonomous MAVs," in Proc. IEEE Int. Conference on Robotics and
Automation, New Orleans, September 2003, vol. 1, pp. 1422-1427.









[11] T. D. Morton, Embedded Microcontrollers, Prentice-Hall, Columbus, 2001.

[12] J. Plew, J. W. Grzywna, M. C. Nechyba, and P. G. Ifju, "Recent Progress in
the Development of On-Board Electronics for Micro Air Vehicles," in Proc.
Florida Conference on the Recent Advances in Robotics, Orlando, August
2004, vol. 17, sec. FP3, pp. 1-6.

[13] J. Plew, "Development of Flight Avionics System for Autonomous MAV
Control," M.S. thesis, University of Florida, December 2004.

[14] J. W. Grzywna, A. Jain, J. Plew, and M. C. N. .ivba, "Rapid Development of
Vision-Based Control for MAVs through a Virtual Flight Testbed," submitted
to IEEE Int. Conf. on Robotics and Automation, April 2005.

[15] M. Quigley, M. A. Goodrich, and R. W. Beard, "Semi-Autonomous Human-
UAV Interfaces for Fixed-Winged Mini-UAVs," to be presented at IEEE Int.
Conf. on Intelligent Robots and Systems, August 2005.

[16] S. M. Ettinger, M. C. N. .Ivba, P. G. Ifju, and M. Waszak, "Vision-guided
Flight Stability and Control for Micro Air Vehicles," in Proceedings IEEE Int.
Conference on Intelligent Robots and S1;1-/. Lausane, October 2002, vol. 3,
pp. 2134-2140.

[17] B. Nadel, \Vo. !. Networking 101," World Wide Web, http://reviews.
cnet.com/4520-3243_7-5021297.html, last accessed on November 15, 2004.

[18] D. A. Forsyth and J. Ponce, Computer Vision: A Modern Approach, Prentice-
Hall, Upper Saddle River, 2003.

[19] A. Kurdila, M. C. Nechyba, R. Lind, P. G. Ifju, W. Dahmen, R. DeVore,
and R. Sharpley, "Vision-based Control of Micro Air Vehicles: Progress and
Problems in Estimation," in Presented at IEEE Int. Conference on Decision
and Control, Atlantis, 2004.

[20] L. Armesto, S. Chroust, M. Vincze, and J. Tornero, \!,I i-rate Fusion with
Vision and Inertial Sensors," in Proc. IEEE Int. Conference on Robotics and
Automation, April 2004, vol. 1, pp. 193-199.

[21] R. C. Dorf and R. H. Bishop, Modern Control S11-1. m- Addison-Wesley,
Menlo Park, 1998.







34

[22] J. W. Grzywna, "A Flight Testbed with Virtual Environment Capabilities
for Developing Autonomous Micro Air Vehicles," M.S. thesis, University of
Florida, December 2004.















BIOGRAPHICAL SKETCH


Uriel Rodriguez was born in Ponce, Puerto Rico, on June 6, 1979. His family

moved to Coral Springs, FL, in 1988. He received his high school diploma from

Marjory Stoneman Douglas High School in Parkland, FL. He then attended the

University of Florida and received his bachelor's degree in computer engineering

in December 2001. While working on his undergraduate work, Uriel was active in

the Student IEEE Branch and was a member of the IEEE SoutheastCon Student

Hardware Team which won 1st Place with its Pong robot. Since then Uriel has

worked in the Machine Intelligence Lab under Dr. Antonio Arroyo, Dr. Michael

Nechyba, and Dr. Eric Schwartz.