First script to experiment with Concurrent calibration estimation

release/4.3a0
cbeall3 2014-06-13 12:17:45 -04:00
parent 9f796565d0
commit 67e0e71802
7 changed files with 141694 additions and 0 deletions

View File

@ -0,0 +1 @@
718.856 718.856 0.0 607.1928 185.2157 0.5371657189

View File

@ -0,0 +1 @@
718.856 718.856 0.0 607.1928 185.2157 0.5371657189

View File

@ -0,0 +1,135 @@
0 1 0 0 0 0 1 0 0 -0 0 1 0 0 0 0 1
1 0.99999 -0.00268679 -0.00354618 6.43221e-05 0.00267957 0.999994 -0.00204036 -0.0073023 0.00355164 0.00203084 0.999992 0.676456 0 0 0 1
2 0.999969 -0.00120771 -0.00772489 -0.0100328 0.00117985 0.999993 -0.003611 -0.0111185 0.00772919 0.00360178 0.999964 1.37125 0 0 0 1
3 0.999931 -0.00128098 -0.0117006 -0.0237327 0.00122052 0.999986 -0.00517227 -0.0136538 0.0117071 0.00515763 0.999918 2.08563 0 0 0 1
4 0.99986 5.79321e-05 -0.0167106 -0.0402272 -0.000155312 0.999983 -0.00582618 -0.0194327 0.01671 0.00582796 0.999843 2.81528 0 0 0 1
5 0.999772 -0.00118366 -0.0213077 -0.0572378 0.0010545 0.999981 -0.00607208 -0.0278191 0.0213145 0.00604822 0.999755 3.56204 0 0 0 1
6 0.999662 0.000544425 -0.0259946 -0.081545 -0.000735472 0.999973 -0.00734051 -0.0358844 0.0259899 0.00735714 0.999635 4.32265 0 0 0 1
7 0.999513 0.0032602 -0.0310324 -0.112137 -0.0035101 0.999962 -0.00800188 -0.0447209 0.0310051 0.00810691 0.999486 5.09668 0 0 0 1
8 0.999361 0.00349173 -0.0355658 -0.143594 -0.00372162 0.999973 -0.00639979 -0.0532611 0.0355425 0.00652807 0.999347 5.88701 0 0 0 1
9 0.999185 0.00268131 -0.040271 -0.176401 -0.0028332 0.999989 -0.00371493 -0.0632884 0.0402606 0.003826 0.999182 6.6897 0 0 0 1
10 0.99903 0.00226305 -0.0439747 -0.211687 -0.00231163 0.999997 -0.00105382 -0.072362 0.0439722 0.00115445 0.999032 7.50361 0 0 0 1
11 0.998896 0.00366482 -0.0468376 -0.254125 -0.00374515 0.999992 -0.00162734 -0.0820263 0.0468312 0.00180096 0.998901 8.32333 0 0 0 1
12 0.998775 0.00304285 -0.0493866 -0.295424 -0.00313866 0.999993 -0.00186268 -0.0885739 0.0493806 0.00201541 0.998778 9.15211 0 0 0 1
13 0.998682 7.09894e-05 -0.0513155 -0.334647 -0.000203775 0.999997 -0.00258241 -0.0938889 0.0513152 0.00258946 0.998679 9.98839 0 0 0 1
14 0.998565 -8.82523e-05 -0.0535542 -0.380835 -9.36659e-06 0.999998 -0.00182255 -0.10173 0.0535542 0.00182044 0.998563 10.832 0 0 0 1
15 0.998481 -0.00146793 -0.0550718 -0.429135 0.0013525 0.999997 -0.00213307 -0.111427 0.0550748 0.00205535 0.99848 11.687 0 0 0 1
16 0.998373 0.000738731 -0.0570218 -0.483426 -0.000993083 0.99999 -0.00443241 -0.122139 0.0570179 0.00448183 0.998363 12.5483 0 0 0 1
17 0.998285 0.00120595 -0.0585258 -0.540056 -0.00162301 0.999974 -0.00707907 -0.132598 0.0585158 0.00716191 0.998261 13.4179 0 0 0 1
18 0.998165 0.00516151 -0.060337 -0.6023 -0.00570195 0.999945 -0.00878826 -0.143753 0.0602883 0.00911617 0.998139 14.2952 0 0 0 1
19 0.998101 0.00610094 -0.0612993 -0.66308 -0.00663017 0.999942 -0.00843386 -0.157854 0.0612443 0.00882427 0.998084 15.1802 0 0 0 1
20 0.998014 0.0052997 -0.0627662 -0.722045 -0.00574767 0.999959 -0.0069587 -0.172847 0.0627268 0.00730564 0.998004 16.074 0 0 0 1
21 0.99792 0.00591748 -0.0641975 -0.78346 -0.00627924 0.999966 -0.00543487 -0.186221 0.0641631 0.00582667 0.997922 16.9738 0 0 0 1
22 0.997857 0.00547694 -0.0651993 -0.845347 -0.00584101 0.999968 -0.00539455 -0.199741 0.0651677 0.00576382 0.997858 17.8786 0 0 0 1
23 0.997737 0.00536917 -0.0670282 -0.908218 -0.00579979 0.999964 -0.0062316 -0.212775 0.0669924 0.00660624 0.997732 18.7877 0 0 0 1
24 0.997663 0.00386695 -0.0682185 -0.971291 -0.00435203 0.999966 -0.00696344 -0.226442 0.0681893 0.00724406 0.997646 19.7046 0 0 0 1
25 0.997629 0.00410637 -0.0687004 -1.03663 -0.00448288 0.999976 -0.00532714 -0.239555 0.0686769 0.00562249 0.997623 20.6257 0 0 0 1
26 0.997617 0.00588773 -0.0687501 -1.10557 -0.0060349 0.99998 -0.00193325 -0.254273 0.0687373 0.00234355 0.997632 21.55 0 0 0 1
27 0.997662 0.00693766 -0.0679906 -1.17297 -0.00682806 0.999975 0.0018442 -0.26563 0.0680017 -0.00137565 0.997684 22.4875 0 0 0 1
28 0.997774 0.00579785 -0.0664343 -1.23728 -0.00550265 0.999974 0.00462554 -0.271962 0.0664594 -0.00424968 0.99778 23.4285 0 0 0 1
29 0.997872 0.00589563 -0.0649408 -1.30214 -0.00556012 0.99997 0.00534586 -0.277922 0.0649704 -0.0049734 0.997875 24.3732 0 0 0 1
30 0.997958 0.00627024 -0.0635595 -1.36462 -0.00612984 0.999978 0.00240374 -0.285335 0.0635732 -0.00200922 0.997975 25.314 0 0 0 1
31 0.998004 0.00714074 -0.0627411 -1.42783 -0.00731158 0.99997 -0.00249375 -0.293171 0.0627215 0.00294751 0.998027 26.2605 0 0 0 1
32 0.99808 0.0063692 -0.0616159 -1.48954 -0.00671918 0.999962 -0.00547459 -0.302321 0.0615787 0.00587809 0.998085 27.2168 0 0 0 1
33 0.99813 0.00376787 -0.0610159 -1.54654 -0.00404632 0.999982 -0.0044408 -0.313516 0.0609981 0.00467938 0.998127 28.1829 0 0 0 1
34 0.998113 0.00193972 -0.0613743 -1.60668 -0.00191171 0.999998 0.000515183 -0.324411 0.0613752 -0.000396881 0.998115 29.1626 0 0 0 1
35 0.99806 -0.0017885 -0.062228 -1.66532 0.00203402 0.99999 0.00388232 -0.335656 0.0622204 -0.00400136 0.998054 30.1428 0 0 0 1
36 0.997945 -0.00917543 -0.0634115 -1.72059 0.00939451 0.999951 0.00315749 -0.343316 0.0633794 -0.00374672 0.997982 31.1244 0 0 0 1
37 0.997825 -0.0112684 -0.0649459 -1.78049 0.011242 0.999937 -0.000771312 -0.350864 0.0649504 3.95099e-05 0.997888 32.1064 0 0 0 1
38 0.997739 -0.0110126 -0.0662983 -1.85007 0.0107254 0.999932 -0.00468596 -0.361068 0.0663454 0.00396429 0.997789 33.0886 0 0 0 1
39 0.997597 -0.00959503 -0.0686163 -1.92119 0.00924037 0.999942 -0.00548426 -0.373466 0.0686649 0.00483704 0.997628 34.0774 0 0 0 1
40 0.99755 -0.0095802 -0.0693031 -1.99331 0.00931271 0.999948 -0.00418184 -0.387047 0.0693396 0.00352619 0.997587 35.0736 0 0 0 1
41 0.997473 -0.00634387 -0.0707596 -2.0707 0.00626661 0.99998 -0.0013139 -0.403858 0.0707665 0.00086716 0.997493 36.0721 0 0 0 1
42 0.99739 -0.00624366 -0.0719343 -2.14553 0.00625582 0.99998 -5.62375e-05 -0.416888 0.0719332 -0.000393917 0.997409 37.0728 0 0 0 1
43 0.997312 -0.00473093 -0.0731254 -2.21909 0.00492848 0.999985 0.00252135 -0.428625 0.0731123 -0.00287497 0.99732 38.0643 0 0 0 1
44 0.997318 -0.00467696 -0.0730348 -2.29215 0.00509473 0.999972 0.00553481 -0.440023 0.0730068 -0.00589206 0.997314 39.0618 0 0 0 1
45 0.997274 0.00138304 -0.0737801 -2.37574 -0.000811217 0.999969 0.00777971 -0.447869 0.0737886 -0.00769865 0.997244 40.0548 0 0 0 1
46 0.997262 0.00149131 -0.0739326 -2.45529 -0.000969511 0.999974 0.00709318 -0.454763 0.0739413 -0.00700208 0.997238 41.0557 0 0 0 1
47 0.997266 0.00175929 -0.0738699 -2.53081 -0.00136899 0.999985 0.00533379 -0.460519 0.0738782 -0.00521809 0.997254 42.0518 0 0 0 1
48 0.997253 0.00408494 -0.0739555 -2.61212 -0.00386552 0.999988 0.00310988 -0.469863 0.0739673 -0.00281546 0.997257 43.0493 0 0 0 1
49 0.997185 0.00365371 -0.0748884 -2.68799 -0.00342799 0.999989 0.00314243 -0.47951 0.0748991 -0.00287687 0.997187 44.0473 0 0 0 1
50 0.997077 0.00181435 -0.0763845 -2.76071 -0.00149292 0.99999 0.00426495 -0.487845 0.0763915 -0.00413845 0.997069 45.0403 0 0 0 1
51 0.997018 0.00246727 -0.0771352 -2.84117 -0.00206285 0.999984 0.00532227 -0.499132 0.0771471 -0.00514727 0.997006 46.0244 0 0 0 1
52 0.996991 0.00504805 -0.0773507 -2.92304 -0.00493379 0.999986 0.00166824 -0.510863 0.0773581 -0.00128158 0.997003 46.994 0 0 0 1
53 0.996911 0.00581773 -0.0783264 -3.00373 -0.00604061 0.999978 -0.00260888 -0.521193 0.0783095 0.00307396 0.996924 47.9551 0 0 0 1
54 0.996846 0.00678413 -0.0790757 -3.08343 -0.00711636 0.999967 -0.00392044 -0.534186 0.0790465 0.00447081 0.996861 48.9236 0 0 0 1
55 0.996843 0.00557268 -0.0792034 -3.16262 -0.00562268 0.999984 -0.000408328 -0.54901 0.0791999 0.000852374 0.996858 49.9005 0 0 0 1
56 0.996831 0.00375007 -0.0794568 -3.23868 -0.00354655 0.99999 0.00270227 -0.563036 0.0794661 -0.0024119 0.996835 50.8752 0 0 0 1
57 0.996805 0.00190455 -0.0798474 -3.31582 -0.00164885 0.999993 0.00326822 -0.574113 0.0798531 -0.00312612 0.996802 51.8394 0 0 0 1
58 0.996782 -0.00124932 -0.0801505 -3.39153 0.00141878 0.999997 0.0020573 -0.586659 0.0801477 -0.00216439 0.996781 52.8005 0 0 0 1
59 0.996745 -0.0038025 -0.0805262 -3.4676 0.0038689 0.999992 0.000668539 -0.59892 0.080523 -0.00097791 0.996752 53.7575 0 0 0 1
60 0.996643 -0.00519016 -0.0817059 -3.54489 0.00535256 0.999984 0.00176869 -0.60864 0.0816955 -0.00220009 0.996655 54.708 0 0 0 1
61 0.996534 -0.0079249 -0.0828082 -3.62139 0.00842977 0.999948 0.00574894 -0.618858 0.0827583 -0.00642707 0.996549 55.6588 0 0 0 1
62 0.996473 -0.00854289 -0.0834829 -3.69959 0.00945654 0.9999 0.0105549 -0.624401 0.0833844 -0.0113071 0.996453 56.6119 0 0 0 1
63 0.996447 -0.00664747 -0.083957 -3.78502 0.00773966 0.99989 0.0126902 -0.629769 0.0838633 -0.0132949 0.996389 57.5607 0 0 0 1
64 0.996335 -0.00522633 -0.0853755 -3.8689 0.00597793 0.999946 0.00855017 -0.636709 0.0853262 -0.0090292 0.996312 58.4941 0 0 0 1
65 0.996221 -0.00343661 -0.0867892 -3.95276 0.00350579 0.999994 0.000644619 -0.644008 0.0867865 -0.000946448 0.996226 59.4131 0 0 0 1
66 0.996144 -0.00149623 -0.0877201 -4.03806 0.00112725 0.99999 -0.00425562 -0.655271 0.0877256 0.00414033 0.996136 60.3236 0 0 0 1
67 0.996055 0.00375138 -0.0886573 -4.12895 -0.00406723 0.999986 -0.00338223 -0.671324 0.0886434 0.00372948 0.996056 61.2274 0 0 0 1
68 0.995922 0.00719305 -0.0899263 -4.21985 -0.0073202 0.999973 -0.00108421 -0.691307 0.089916 0.00173807 0.995948 62.125 0 0 0 1
69 0.99582 0.00967277 -0.0908194 -4.30702 -0.00966905 0.999953 0.000481019 -0.708494 0.0908198 0.000399128 0.995867 63.0134 0 0 0 1
70 0.995713 0.0102896 -0.0919182 -4.39131 -0.0103098 0.999947 0.000255248 -0.721276 0.091916 0.000693502 0.995767 63.8776 0 0 0 1
71 0.99554 0.0119225 -0.0935844 -4.477 -0.0118725 0.999929 0.00109156 -0.734766 0.0935908 2.43836e-05 0.995611 64.7307 0 0 0 1
72 0.995397 0.0126524 -0.0950024 -4.56121 -0.0125521 0.99992 0.00165348 -0.749039 0.0950157 -0.000453392 0.995476 65.5703 0 0 0 1
73 0.995256 0.0126635 -0.0964665 -4.64297 -0.0125254 0.999919 0.00203772 -0.761909 0.0964846 -0.00081977 0.995334 66.3938 0 0 0 1
74 0.995133 0.0127023 -0.0977168 -4.72623 -0.0124698 0.999918 0.00298947 -0.7711 0.0977468 -0.00175641 0.99521 67.2017 0 0 0 1
75 0.994948 0.015548 -0.0991814 -4.81287 -0.0150604 0.999871 0.00566291 -0.780301 0.0992566 -0.0041406 0.995053 67.9995 0 0 0 1
76 0.994794 0.0171065 -0.100462 -4.90076 -0.0162095 0.999821 0.009738 -0.788037 0.100611 -0.00805885 0.994893 68.7922 0 0 0 1
77 0.994568 0.0201446 -0.102122 -4.98771 -0.0190681 0.999752 0.0115061 -0.794955 0.102328 -0.00949629 0.994705 69.5653 0 0 0 1
78 0.99437 0.0229894 -0.10344 -5.07707 -0.0223435 0.999723 0.00739861 -0.804841 0.103581 -0.00504575 0.994608 70.3178 0 0 0 1
79 0.99423 0.0228747 -0.1048 -5.16223 -0.0229899 0.999736 0.000108822 -0.81531 0.104774 0.00230114 0.994493 71.0475 0 0 0 1
80 0.994077 0.0219938 -0.106431 -5.24094 -0.0227304 0.999725 -0.00571252 -0.825441 0.106276 0.00809789 0.994304 71.7584 0 0 0 1
81 0.994023 0.0228054 -0.106762 -5.32437 -0.0236929 0.999694 -0.00705161 -0.831667 0.106569 0.00953897 0.99426 72.4548 0 0 0 1
82 0.99386 0.0255543 -0.107653 -5.40648 -0.0260808 0.999654 -0.00348505 -0.846106 0.107527 0.00627133 0.994182 73.1337 0 0 0 1
83 0.993702 0.0257681 -0.109048 -5.48436 -0.02605 0.99966 -0.00116096 -0.865059 0.108981 0.00399435 0.994036 73.7942 0 0 0 1
84 0.99367 0.0225468 -0.110051 -5.5561 -0.0231761 0.999722 -0.00444219 -0.879535 0.10992 0.00696462 0.993916 74.4324 0 0 0 1
85 0.993802 0.0143509 -0.110234 -5.61528 -0.0155198 0.999832 -0.00975281 -0.89282 0.110075 0.0114032 0.993858 75.0504 0 0 0 1
86 0.993949 0.0102 -0.10937 -5.67796 -0.0118817 0.999821 -0.0147354 -0.904058 0.1092 0.0159457 0.993892 75.6535 0 0 0 1
87 0.994244 0.0126451 -0.106395 -5.74524 -0.014328 0.999784 -0.0150673 -0.916949 0.106181 0.016505 0.99421 76.2455 0 0 0 1
88 0.994592 0.0175824 -0.102356 -5.81375 -0.0189231 0.999747 -0.0121417 -0.930972 0.102117 0.0140129 0.994674 76.8222 0 0 0 1
89 0.995077 0.0159295 -0.0978149 -5.86699 -0.0169123 0.999814 -0.00922648 -0.942909 0.0976498 0.0108353 0.995162 77.3862 0 0 0 1
90 0.995665 0.0122046 -0.0922106 -5.90659 -0.0127358 0.999906 -0.00517412 -0.954487 0.0921387 0.00632606 0.995726 77.9355 0 0 0 1
91 0.996426 0.00781104 -0.084105 -5.94257 -0.00798227 0.999967 -0.0016998 -0.970792 0.0840889 0.00236507 0.996455 78.4692 0 0 0 1
92 0.997233 0.0114593 -0.0734453 -5.98049 -0.0116369 0.99993 -0.00198965 -0.981019 0.0734174 0.00283882 0.997297 78.9883 0 0 0 1
93 0.998165 0.0165636 -0.0582361 -6.0154 -0.0167456 0.999856 -0.0026387 -0.99003 0.058184 0.00360906 0.998299 79.4952 0 0 0 1
95 0.999635 0.0200255 -0.0181151 -6.02981 -0.0200623 0.999797 -0.00185529 -1.00502 0.0180742 0.00221804 0.999834 80.4727 0 0 0 1
97 0.999162 0.015548 0.037857 -5.96801 -0.0155684 0.999879 0.000243918 -1.02389 -0.0378486 -0.000833085 0.999283 81.4025 0 0 0 1
99 0.993959 0.0151454 0.108698 -5.84553 -0.0154328 0.999879 0.0018028 -1.04109 -0.108657 -0.00346942 0.994073 82.2952 0 0 0 1
101 0.980499 0.0151504 0.195937 -5.64466 -0.0157106 0.999876 0.00130478 -1.05761 -0.195893 -0.00435763 0.980616 83.1489 0 0 0 1
103 0.954186 0.0182833 0.298656 -5.36588 -0.0179595 0.999831 -0.00382887 -1.08348 -0.298675 -0.00171027 0.954353 83.9397 0 0 0 1
105 0.910736 0.0194893 0.412529 -4.99648 -0.0175815 0.99981 -0.00842014 -1.10057 -0.412615 0.000415655 0.910905 84.6633 0 0 0 1
107 0.848724 0.0183908 0.528517 -4.54701 -0.0135972 0.999824 -0.0129557 -1.12003 -0.528662 0.00380946 0.848824 85.2983 0 0 0 1
109 0.772259 0.0170098 0.63508 -4.0183 -0.0106749 0.999848 -0.0137989 -1.14244 -0.635218 0.00387688 0.772323 85.8601 0 0 0 1
111 0.684256 0.0156411 0.729074 -3.42903 -0.0102231 0.999877 -0.0118561 -1.16079 -0.72917 0.000659179 0.684332 86.3474 0 0 0 1
113 0.590745 0.011826 0.806772 -2.77931 -0.000173089 0.999894 -0.0145301 -1.17886 -0.806858 0.00844396 0.590684 86.7443 0 0 0 1
115 0.496173 0.0169181 0.868059 -2.10955 -0.00504039 0.999849 -0.0166057 -1.20339 -0.868209 0.00386392 0.496183 87.0847 0 0 0 1
117 0.408192 0.0165355 0.912746 -1.40862 0.00231553 0.999814 -0.0191484 -1.2249 -0.912893 0.00992974 0.408078 87.3396 0 0 0 1
119 0.333443 0.00543386 0.942754 -0.671521 0.0223493 0.999657 -0.0136666 -1.23662 -0.942505 0.025627 0.333208 87.522 0 0 0 1
121 0.269054 0.0173163 0.962969 0.0638526 0.00961829 0.99974 -0.0206648 -1.26307 -0.963077 0.0148221 0.268818 87.7199 0 0 0 1
123 0.214897 0.0233915 0.976357 0.843046 0.00677025 0.999653 -0.0254398 -1.30009 -0.976613 0.0120771 0.214664 87.8763 0 0 0 1
125 0.171479 0.031054 0.984698 1.66216 0.0212619 0.999154 -0.0352125 -1.3179 -0.984958 0.0269747 0.170674 87.9743 0 0 0 1
127 0.134011 0.0386308 0.990227 2.52547 0.0207141 0.998912 -0.041773 -1.34147 -0.990763 0.0261097 0.133065 88.0809 0 0 0 1
129 0.10418 0.0310179 0.994075 3.44652 0.0195614 0.999256 -0.0332297 -1.39013 -0.994366 0.0229074 0.103496 88.1692 0 0 0 1
131 0.0794366 0.027788 0.996453 4.42556 0.0261822 0.999208 -0.0299521 -1.42776 -0.996496 0.0284686 0.0786462 88.2295 0 0 0 1
132 0.0693462 0.028443 0.997187 4.93885 0.0294969 0.999098 -0.0305488 -1.44582 -0.997156 0.0315324 0.0684447 88.2553 0 0 0 1
133 0.0615414 0.0290168 0.997683 5.46907 0.0316982 0.999016 -0.0310108 -1.46406 -0.997601 0.0335332 0.0605611 88.2814 0 0 0 1
134 0.0559347 0.029371 0.998002 6.0151 0.0334765 0.99895 -0.0312751 -1.48373 -0.997873 0.035159 0.0548927 88.307 0 0 0 1
135 0.0504312 0.0304374 0.998264 6.58025 0.0349281 0.99887 -0.0322204 -1.50267 -0.998117 0.0364923 0.0493112 88.3306 0 0 0 1
136 0.0445067 0.0311103 0.998525 7.16082 0.0355578 0.998832 -0.0327048 -1.52353 -0.998376 0.0369609 0.0433485 88.3531 0 0 0 1
137 0.040243 0.0311989 0.998703 7.76375 0.0381603 0.998735 -0.0327376 -1.54487 -0.998461 0.0394283 0.0390016 88.3716 0 0 0 1
138 0.0373982 0.0312027 0.998813 8.38568 0.0397152 0.998676 -0.0326855 -1.56772 -0.998511 0.0408905 0.0361095 88.3901 0 0 0 1
139 0.0343726 0.0307634 0.998936 9.02449 0.0406913 0.998654 -0.0321549 -1.59059 -0.99858 0.0417533 0.0330745 88.4092 0 0 0 1
140 0.0320861 0.0302694 0.999027 9.68038 0.0427798 0.998584 -0.03163 -1.61442 -0.998569 0.043753 0.0307457 88.4263 0 0 0 1
141 0.0316452 0.0299561 0.99905 10.3542 0.0473602 0.998383 -0.0314363 -1.63856 -0.998376 0.04831 0.0301753 88.4381 0 0 0 1
142 0.0327723 0.029714 0.999021 11.0457 0.0507142 0.998221 -0.0313539 -1.66282 -0.998175 0.0516921 0.031207 88.4556 0 0 0 1
143 0.0353027 0.0297602 0.998933 11.7546 0.0522842 0.998133 -0.0315841 -1.68678 -0.998008 0.0533435 0.0336808 88.4781 0 0 0 1
144 0.0392372 0.0297502 0.998787 12.4771 0.0547241 0.997993 -0.0318763 -1.71289 -0.99773 0.0559084 0.0375304 88.5062 0 0 0 1
145 0.0437096 0.0293188 0.998614 13.219 0.0550685 0.997979 -0.0317105 -1.73922 -0.997525 0.0563782 0.0420067 88.5387 0 0 0 1
146 0.0477725 0.0278103 0.998471 13.9764 0.0564652 0.997939 -0.0304971 -1.76499 -0.997261 0.0578358 0.0461037 88.5751 0 0 0 1
147 0.0518486 0.0263145 0.998308 14.7472 0.0562418 0.997989 -0.0292271 -1.79222 -0.99707 0.057662 0.0502644 88.6178 0 0 0 1
148 0.0560658 0.0242863 0.998132 15.5313 0.0531494 0.998214 -0.0272738 -1.82056 -0.997011 0.0545792 0.0546748 88.6693 0 0 0 1
149 0.0600218 0.0233355 0.997924 16.3271 0.0522059 0.998285 -0.0264839 -1.84733 -0.996831 0.0536871 0.0587006 88.7243 0 0 0 1
150 0.0641513 0.0243795 0.997642 17.1258 0.0492204 0.998408 -0.0275632 -1.87761 -0.996726 0.0508726 0.0628492 88.7821 0 0 0 1
151 0.0672583 0.028483 0.997329 17.929 0.0470717 0.998389 -0.0316877 -1.91204 -0.996625 0.0490772 0.0658092 88.842 0 0 0 1
152 0.0688453 0.0337446 0.997056 18.7357 0.0413971 0.99847 -0.0366509 -1.9468 -0.996768 0.0437985 0.067343 88.9041 0 0 0 1
153 0.0686545 0.0370247 0.996953 19.5482 0.0387033 0.99846 -0.0397459 -1.98038 -0.996889 0.0413142 0.0671158 88.9665 0 0 0 1

View File

@ -0,0 +1,77 @@
0 1 0 0 0 0 1 0 0 -0 0 1 0 0 0 0 1
1 0.99999 -0.00268679 -0.00354618 6.43221e-05 0.00267957 0.999994 -0.00204036 -0.0073023 0.00355164 0.00203084 0.999992 0.676456 0 0 0 1
2 0.999969 -0.00120771 -0.00772489 -0.0100328 0.00117985 0.999993 -0.003611 -0.0111185 0.00772919 0.00360178 0.999964 1.37125 0 0 0 1
3 0.999931 -0.00128098 -0.0117006 -0.0237327 0.00122052 0.999986 -0.00517227 -0.0136538 0.0117071 0.00515763 0.999918 2.08563 0 0 0 1
4 0.99986 5.79321e-05 -0.0167106 -0.0402272 -0.000155312 0.999983 -0.00582618 -0.0194327 0.01671 0.00582796 0.999843 2.81528 0 0 0 1
5 0.999772 -0.00118366 -0.0213077 -0.0572378 0.0010545 0.999981 -0.00607208 -0.0278191 0.0213145 0.00604822 0.999755 3.56204 0 0 0 1
6 0.999662 0.000544425 -0.0259946 -0.081545 -0.000735472 0.999973 -0.00734051 -0.0358844 0.0259899 0.00735714 0.999635 4.32265 0 0 0 1
7 0.999513 0.0032602 -0.0310324 -0.112137 -0.0035101 0.999962 -0.00800188 -0.0447209 0.0310051 0.00810691 0.999486 5.09668 0 0 0 1
8 0.999361 0.00349173 -0.0355658 -0.143594 -0.00372162 0.999973 -0.00639979 -0.0532611 0.0355425 0.00652807 0.999347 5.88701 0 0 0 1
9 0.999185 0.00268131 -0.040271 -0.176401 -0.0028332 0.999989 -0.00371493 -0.0632884 0.0402606 0.003826 0.999182 6.6897 0 0 0 1
10 0.99903 0.00226305 -0.0439747 -0.211687 -0.00231163 0.999997 -0.00105382 -0.072362 0.0439722 0.00115445 0.999032 7.50361 0 0 0 1
11 0.998896 0.00366482 -0.0468376 -0.254125 -0.00374515 0.999992 -0.00162734 -0.0820263 0.0468312 0.00180096 0.998901 8.32333 0 0 0 1
12 0.998775 0.00304285 -0.0493866 -0.295424 -0.00313866 0.999993 -0.00186268 -0.0885739 0.0493806 0.00201541 0.998778 9.15211 0 0 0 1
13 0.998682 7.09894e-05 -0.0513155 -0.334647 -0.000203775 0.999997 -0.00258241 -0.0938889 0.0513152 0.00258946 0.998679 9.98839 0 0 0 1
14 0.998565 -8.82523e-05 -0.0535542 -0.380835 -9.36659e-06 0.999998 -0.00182255 -0.10173 0.0535542 0.00182044 0.998563 10.832 0 0 0 1
15 0.998481 -0.00146793 -0.0550718 -0.429135 0.0013525 0.999997 -0.00213307 -0.111427 0.0550748 0.00205535 0.99848 11.687 0 0 0 1
16 0.998373 0.000738731 -0.0570218 -0.483426 -0.000993083 0.99999 -0.00443241 -0.122139 0.0570179 0.00448183 0.998363 12.5483 0 0 0 1
17 0.998285 0.00120595 -0.0585258 -0.540056 -0.00162301 0.999974 -0.00707907 -0.132598 0.0585158 0.00716191 0.998261 13.4179 0 0 0 1
18 0.998165 0.00516151 -0.060337 -0.6023 -0.00570195 0.999945 -0.00878826 -0.143753 0.0602883 0.00911617 0.998139 14.2952 0 0 0 1
19 0.998101 0.00610094 -0.0612993 -0.66308 -0.00663017 0.999942 -0.00843386 -0.157854 0.0612443 0.00882427 0.998084 15.1802 0 0 0 1
20 0.998014 0.0052997 -0.0627662 -0.722045 -0.00574767 0.999959 -0.0069587 -0.172847 0.0627268 0.00730564 0.998004 16.074 0 0 0 1
21 0.99792 0.00591748 -0.0641975 -0.78346 -0.00627924 0.999966 -0.00543487 -0.186221 0.0641631 0.00582667 0.997922 16.9738 0 0 0 1
22 0.997857 0.00547694 -0.0651993 -0.845347 -0.00584101 0.999968 -0.00539455 -0.199741 0.0651677 0.00576382 0.997858 17.8786 0 0 0 1
23 0.997737 0.00536917 -0.0670282 -0.908218 -0.00579979 0.999964 -0.0062316 -0.212775 0.0669924 0.00660624 0.997732 18.7877 0 0 0 1
24 0.997663 0.00386695 -0.0682185 -0.971291 -0.00435203 0.999966 -0.00696344 -0.226442 0.0681893 0.00724406 0.997646 19.7046 0 0 0 1
25 0.997629 0.00410637 -0.0687004 -1.03663 -0.00448288 0.999976 -0.00532714 -0.239555 0.0686769 0.00562249 0.997623 20.6257 0 0 0 1
26 0.997617 0.00588773 -0.0687501 -1.10557 -0.0060349 0.99998 -0.00193325 -0.254273 0.0687373 0.00234355 0.997632 21.55 0 0 0 1
27 0.997662 0.00693766 -0.0679906 -1.17297 -0.00682806 0.999975 0.0018442 -0.26563 0.0680017 -0.00137565 0.997684 22.4875 0 0 0 1
28 0.997774 0.00579785 -0.0664343 -1.23728 -0.00550265 0.999974 0.00462554 -0.271962 0.0664594 -0.00424968 0.99778 23.4285 0 0 0 1
29 0.997872 0.00589563 -0.0649408 -1.30214 -0.00556012 0.99997 0.00534586 -0.277922 0.0649704 -0.0049734 0.997875 24.3732 0 0 0 1
30 0.997958 0.00627024 -0.0635595 -1.36462 -0.00612984 0.999978 0.00240374 -0.285335 0.0635732 -0.00200922 0.997975 25.314 0 0 0 1
31 0.998004 0.00714074 -0.0627411 -1.42783 -0.00731158 0.99997 -0.00249375 -0.293171 0.0627215 0.00294751 0.998027 26.2605 0 0 0 1
32 0.99808 0.0063692 -0.0616159 -1.48954 -0.00671918 0.999962 -0.00547459 -0.302321 0.0615787 0.00587809 0.998085 27.2168 0 0 0 1
33 0.99813 0.00376787 -0.0610159 -1.54654 -0.00404632 0.999982 -0.0044408 -0.313516 0.0609981 0.00467938 0.998127 28.1829 0 0 0 1
34 0.998113 0.00193972 -0.0613743 -1.60668 -0.00191171 0.999998 0.000515183 -0.324411 0.0613752 -0.000396881 0.998115 29.1626 0 0 0 1
35 0.99806 -0.0017885 -0.062228 -1.66532 0.00203402 0.99999 0.00388232 -0.335656 0.0622204 -0.00400136 0.998054 30.1428 0 0 0 1
36 0.997945 -0.00917543 -0.0634115 -1.72059 0.00939451 0.999951 0.00315749 -0.343316 0.0633794 -0.00374672 0.997982 31.1244 0 0 0 1
37 0.997825 -0.0112684 -0.0649459 -1.78049 0.011242 0.999937 -0.000771312 -0.350864 0.0649504 3.95099e-05 0.997888 32.1064 0 0 0 1
38 0.997739 -0.0110126 -0.0662983 -1.85007 0.0107254 0.999932 -0.00468596 -0.361068 0.0663454 0.00396429 0.997789 33.0886 0 0 0 1
39 0.997597 -0.00959503 -0.0686163 -1.92119 0.00924037 0.999942 -0.00548426 -0.373466 0.0686649 0.00483704 0.997628 34.0774 0 0 0 1
40 0.99755 -0.0095802 -0.0693031 -1.99331 0.00931271 0.999948 -0.00418184 -0.387047 0.0693396 0.00352619 0.997587 35.0736 0 0 0 1
41 0.997473 -0.00634387 -0.0707596 -2.0707 0.00626661 0.99998 -0.0013139 -0.403858 0.0707665 0.00086716 0.997493 36.0721 0 0 0 1
42 0.99739 -0.00624366 -0.0719343 -2.14553 0.00625582 0.99998 -5.62375e-05 -0.416888 0.0719332 -0.000393917 0.997409 37.0728 0 0 0 1
43 0.997312 -0.00473093 -0.0731254 -2.21909 0.00492848 0.999985 0.00252135 -0.428625 0.0731123 -0.00287497 0.99732 38.0643 0 0 0 1
44 0.997318 -0.00467696 -0.0730348 -2.29215 0.00509473 0.999972 0.00553481 -0.440023 0.0730068 -0.00589206 0.997314 39.0618 0 0 0 1
45 0.997274 0.00138304 -0.0737801 -2.37574 -0.000811217 0.999969 0.00777971 -0.447869 0.0737886 -0.00769865 0.997244 40.0548 0 0 0 1
46 0.997262 0.00149131 -0.0739326 -2.45529 -0.000969511 0.999974 0.00709318 -0.454763 0.0739413 -0.00700208 0.997238 41.0557 0 0 0 1
47 0.997266 0.00175929 -0.0738699 -2.53081 -0.00136899 0.999985 0.00533379 -0.460519 0.0738782 -0.00521809 0.997254 42.0518 0 0 0 1
48 0.997253 0.00408494 -0.0739555 -2.61212 -0.00386552 0.999988 0.00310988 -0.469863 0.0739673 -0.00281546 0.997257 43.0493 0 0 0 1
49 0.997185 0.00365371 -0.0748884 -2.68799 -0.00342799 0.999989 0.00314243 -0.47951 0.0748991 -0.00287687 0.997187 44.0473 0 0 0 1
50 0.997077 0.00181435 -0.0763845 -2.76071 -0.00149292 0.99999 0.00426495 -0.487845 0.0763915 -0.00413845 0.997069 45.0403 0 0 0 1
51 0.997018 0.00246727 -0.0771352 -2.84117 -0.00206285 0.999984 0.00532227 -0.499132 0.0771471 -0.00514727 0.997006 46.0244 0 0 0 1
52 0.996991 0.00504805 -0.0773507 -2.92304 -0.00493379 0.999986 0.00166824 -0.510863 0.0773581 -0.00128158 0.997003 46.994 0 0 0 1
53 0.996911 0.00581773 -0.0783264 -3.00373 -0.00604061 0.999978 -0.00260888 -0.521193 0.0783095 0.00307396 0.996924 47.9551 0 0 0 1
54 0.996846 0.00678413 -0.0790757 -3.08343 -0.00711636 0.999967 -0.00392044 -0.534186 0.0790465 0.00447081 0.996861 48.9236 0 0 0 1
55 0.996843 0.00557268 -0.0792034 -3.16262 -0.00562268 0.999984 -0.000408328 -0.54901 0.0791999 0.000852374 0.996858 49.9005 0 0 0 1
56 0.996831 0.00375007 -0.0794568 -3.23868 -0.00354655 0.99999 0.00270227 -0.563036 0.0794661 -0.0024119 0.996835 50.8752 0 0 0 1
57 0.996805 0.00190455 -0.0798474 -3.31582 -0.00164885 0.999993 0.00326822 -0.574113 0.0798531 -0.00312612 0.996802 51.8394 0 0 0 1
58 0.996782 -0.00124932 -0.0801505 -3.39153 0.00141878 0.999997 0.0020573 -0.586659 0.0801477 -0.00216439 0.996781 52.8005 0 0 0 1
59 0.996745 -0.0038025 -0.0805262 -3.4676 0.0038689 0.999992 0.000668539 -0.59892 0.080523 -0.00097791 0.996752 53.7575 0 0 0 1
60 0.996643 -0.00519016 -0.0817059 -3.54489 0.00535256 0.999984 0.00176869 -0.60864 0.0816955 -0.00220009 0.996655 54.708 0 0 0 1
61 0.996534 -0.0079249 -0.0828082 -3.62139 0.00842977 0.999948 0.00574894 -0.618858 0.0827583 -0.00642707 0.996549 55.6588 0 0 0 1
62 0.996473 -0.00854289 -0.0834829 -3.69959 0.00945654 0.9999 0.0105549 -0.624401 0.0833844 -0.0113071 0.996453 56.6119 0 0 0 1
63 0.996447 -0.00664747 -0.083957 -3.78502 0.00773966 0.99989 0.0126902 -0.629769 0.0838633 -0.0132949 0.996389 57.5607 0 0 0 1
64 0.996335 -0.00522633 -0.0853755 -3.8689 0.00597793 0.999946 0.00855017 -0.636709 0.0853262 -0.0090292 0.996312 58.4941 0 0 0 1
65 0.996221 -0.00343661 -0.0867892 -3.95276 0.00350579 0.999994 0.000644619 -0.644008 0.0867865 -0.000946448 0.996226 59.4131 0 0 0 1
66 0.996144 -0.00149623 -0.0877201 -4.03806 0.00112725 0.99999 -0.00425562 -0.655271 0.0877256 0.00414033 0.996136 60.3236 0 0 0 1
67 0.996055 0.00375138 -0.0886573 -4.12895 -0.00406723 0.999986 -0.00338223 -0.671324 0.0886434 0.00372948 0.996056 61.2274 0 0 0 1
68 0.995922 0.00719305 -0.0899263 -4.21985 -0.0073202 0.999973 -0.00108421 -0.691307 0.089916 0.00173807 0.995948 62.125 0 0 0 1
69 0.99582 0.00967277 -0.0908194 -4.30702 -0.00966905 0.999953 0.000481019 -0.708494 0.0908198 0.000399128 0.995867 63.0134 0 0 0 1
70 0.995713 0.0102896 -0.0919182 -4.39131 -0.0103098 0.999947 0.000255248 -0.721276 0.091916 0.000693502 0.995767 63.8776 0 0 0 1
71 0.99554 0.0119225 -0.0935844 -4.477 -0.0118725 0.999929 0.00109156 -0.734766 0.0935908 2.43836e-05 0.995611 64.7307 0 0 0 1
72 0.995397 0.0126524 -0.0950024 -4.56121 -0.0125521 0.99992 0.00165348 -0.749039 0.0950157 -0.000453392 0.995476 65.5703 0 0 0 1
73 0.995256 0.0126635 -0.0964665 -4.64297 -0.0125254 0.999919 0.00203772 -0.761909 0.0964846 -0.00081977 0.995334 66.3938 0 0 0 1
74 0.995133 0.0127023 -0.0977168 -4.72623 -0.0124698 0.999918 0.00298947 -0.7711 0.0977468 -0.00175641 0.99521 67.2017 0 0 0 1
75 0.994948 0.015548 -0.0991814 -4.81287 -0.0150604 0.999871 0.00566291 -0.780301 0.0992566 -0.0041406 0.995053 67.9995 0 0 0 1
76 0.994794 0.0171065 -0.100462 -4.90076 -0.0162095 0.999821 0.009738 -0.788037 0.100611 -0.00805885 0.994893 68.7922 0 0 0 1

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -0,0 +1,155 @@
/* ----------------------------------------------------------------------------
* GTSAM Copyright 2010, Georgia Tech Research Corporation,
* Atlanta, Georgia 30332-0415
* All Rights Reserved
* Authors: Frank Dellaert, et al. (see THANKS for the full author list)
* See LICENSE for the license information
* -------------------------------------------------------------------------- */
/**
* @file ConcurrentCalibration.cpp
* @brief First step towards estimating monocular calibration in concurrent
* filter/smoother framework. To start with, just batch LM.
* @date June 11, 2014
* @author Chris Beall
*/
#include <gtsam/geometry/Pose3.h>
#include <gtsam/nonlinear/Values.h>
#include <gtsam/nonlinear/NonlinearEquality.h>
#include <gtsam/nonlinear/NonlinearFactorGraph.h>
#include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h>
#include <gtsam/inference/Symbol.h>
#include <gtsam/slam/ProjectionFactor.h>
#include <gtsam/slam/GeneralSFMFactor.h>
#include <gtsam/slam/PriorFactor.h>
#include <gtsam/slam/dataset.h>
#include <string>
#include <fstream>
#include <iostream>
#include <boost/lexical_cast.hpp>
using namespace std;
using namespace gtsam;
int main(int argc, char** argv){
Values initial_estimate;
NonlinearFactorGraph graph;
const noiseModel::Isotropic::shared_ptr model = noiseModel::Isotropic::Sigma(2,1);
string calibration_loc = findExampleDataFile("VO_calibration00s.txt");
string pose_loc = findExampleDataFile("VO_camera_poses00s.txt");
string factor_loc = findExampleDataFile("VO_stereo_factors00s.txt");
//read camera calibration info from file
// focal lengths fx, fy, skew s, principal point u0, v0, baseline b
double fx, fy, s, u0, v0, b;
ifstream calibration_file(calibration_loc.c_str());
cout << "Reading calibration info" << endl;
calibration_file >> fx >> fy >> s >> u0 >> v0 >> b;
//create stereo camera calibration object
const Cal3_S2::shared_ptr K(new Cal3_S2(fx,fy,s,u0,v0));
const Cal3_S2::shared_ptr noisy_K(new Cal3_S2(fx*1.2,fy*1.2,s,u0-10,v0+10));
initial_estimate.insert(Symbol('K', 0), *noisy_K);
noiseModel::Diagonal::shared_ptr calNoise = noiseModel::Diagonal::Sigmas((Vector(5) << 500, 500, 1e-5, 100, 100));
graph.push_back(PriorFactor<Cal3_S2>(Symbol('K', 0), *noisy_K, calNoise));
ifstream pose_file(pose_loc.c_str());
cout << "Reading camera poses" << endl;
int pose_id;
MatrixRowMajor m(4,4);
//read camera pose parameters and use to make initial estimates of camera poses
while (pose_file >> pose_id) {
for (int i = 0; i < 16; i++) {
pose_file >> m.data()[i];
}
initial_estimate.insert(Symbol('x', pose_id), Pose3(m));
}
noiseModel::Isotropic::shared_ptr poseNoise = noiseModel::Isotropic::Sigma(6, 0.01);
graph.push_back(PriorFactor<Pose3>(Symbol('x', pose_id), Pose3(m), poseNoise));
// camera and landmark keys
size_t x, l;
// pixel coordinates uL, uR, v (same for left/right images due to rectification)
// landmark coordinates X, Y, Z in camera frame, resulting from triangulation
double uL, uR, v, X, Y, Z;
ifstream factor_file(factor_loc.c_str());
cout << "Reading stereo factors" << endl;
//read stereo measurement details from file and use to create and add GenericStereoFactor objects to the graph representation
while (factor_file >> x >> l >> uL >> uR >> v >> X >> Y >> Z) {
// graph.push_back( GenericStereoFactor<Pose3, Point3>(StereoPoint2(uL, uR, v), model, Symbol('x', x), Symbol('l', l), K));
graph.push_back(GeneralSFMFactor2<Cal3_S2>(Point2(uL,v), model, Symbol('x', x), Symbol('l', l), Symbol('K', 0)));
//if the landmark variable included in this factor has not yet been added to the initial variable value estimate, add it
if (!initial_estimate.exists(Symbol('l', l))) {
Pose3 camPose = initial_estimate.at<Pose3>(Symbol('x', x));
//transform_from() transforms the input Point3 from the camera pose space, camPose, to the global space
Point3 worldPoint = camPose.transform_from(Point3(X, Y, Z));
initial_estimate.insert(Symbol('l', l), worldPoint);
}
}
Pose3 first_pose = initial_estimate.at<Pose3>(Symbol('x',1));
//constrain the first pose such that it cannot change from its original value during optimization
// NOTE: NonlinearEquality forces the optimizer to use QR rather than Cholesky
// QR is much slower than Cholesky, but numerically more stable
graph.push_back(NonlinearEquality<Pose3>(Symbol('x',1),first_pose));
cout << "Optimizing" << endl;
LevenbergMarquardtParams params;
params.verbosityLM = LevenbergMarquardtParams::TRYLAMBDA;
params.verbosity = NonlinearOptimizerParams::ERROR;
//create Levenberg-Marquardt optimizer to optimize the factor graph
LevenbergMarquardtOptimizer optimizer = LevenbergMarquardtOptimizer(graph, initial_estimate,params);
// Values result = optimizer.optimize();
string K_values_file = "K_values.txt";
ofstream stream_K(K_values_file.c_str());
double currentError;
stream_K << optimizer.iterations() << " " << optimizer.values().at<Cal3_S2>(Symbol('K',0)).vector().transpose() << endl;
// Iterative loop
do {
// Do next iteration
currentError = optimizer.error();
optimizer.iterate();
stream_K << optimizer.iterations() << " " << optimizer.values().at<Cal3_S2>(Symbol('K',0)).vector().transpose() << endl;
if(params.verbosity >= NonlinearOptimizerParams::ERROR) cout << "newError: " << optimizer.error() << endl;
} while(optimizer.iterations() < params.maxIterations &&
!checkConvergence(params.relativeErrorTol, params.absoluteErrorTol,
params.errorTol, currentError, optimizer.error(), params.verbosity));
Values result = optimizer.values();
cout << "Final result sample:" << endl;
Values pose_values = result.filter<Pose3>();
pose_values.print("Final camera poses:\n");
Values(result.filter<Cal3_S2>()).print("Final K\n");
noisy_K->print("Initial noisy K\n");
K->print("Initial correct K\n");
return 0;
}