SLAMの概要と画像を用いた3Dモデリングの基礎

1. 1@2 A / 807 5 / 1 3

2. 1@2 A / 807 5 /

3. 1@2 A / 807 5 / 2 3 • 0 Y 3 • Y 2 3 • ) • ) P • - 2 3 U • 2 ( 1 P 3 2 • 210 C D1 C A 数カ月後シーンの変化復旧・復興の過程１年目２年目年目 ) 6 I A5 M2 3 8

4. 1@2 A / 807 5 / A 4 Deep Matching ! Outlier Rejection 5 Point Algorithm !′ DeepFlow #$# #′ DOF-CDN #% - s s R RB I FR L D N GS- L D c C

5. 1@2 A / 807 5 /

6. 1@2 A / 807 5 / 0 M S 6 L6 30 3

7. 1@2 A / 807 5 / AC I = H . A IA / A ? b [h ge af i cP 2 ]l d SU nmRZU T / 7 =A =I , / 2 II H LLL N I := C L I K A M0 /A G H I = H

8. 1@2 A / 807 5 / c l f o gm • • L a M o gm 8 e • A • • i L • d nVSM ps tu VS

9. 1@2 A / 807 5 / LRU JM 40 1 W • 0 I V RU IPS I7 , 7 , 40 1 G7 , 40 1 C - ECC 2 A: CC 0 D D 9 1A C 0 40 1 G A F 40 1 CE 40 1 2 J. Engel and T. Sch¨ops and D. Cremers G04 40 1 0 4: :D 1A A:E 40 1G Here, σ2 f is the signal variance and l is a length scale that de- termines the strength of correlation between points. Both pa- rameters control the smoothness of the functions estimated by a GP. As can be seen in (2), the covariance between function values decreases with the distance between their corresponding input values. Since we do not have access to the function values, but only noisy observations thereof, it is necessary to represent the corresponding covariance function for noisy observations: cov (yr, ys) = k(xi, xj) + σ2 nδrs . (3) Here σ2 n is the Gaussian observation noise variance and δrs is one if r = s and zero otherwise. For an entire set of input values X, the covariance over the corresponding observations Y becomes cov (Y) = K + σ2 nI, (4) where K is the n × n covariance matrix of the input values, Figure 1: GP model for signal strength of one access point. Black dots indicate training data along the ground truth path. 3.2 Application to Signal Strength Modeling

10. 1@2 A / 807 5 / pnt KD G Cof • lcIof ta hI GF KL G Ce r RA- 0 ,BS M cA B 1 0 , 1 0 , 1 0 , 0 , AO 0 ,B PcA B a bam A 0 0 ,B

11. 1@2 A / 807 5 / tn L e oi m RuST f e kd arb cg V s ptn l O s p My 12 0 My 0 2 0GD D I12 0 A 1C:A BGD : 0 E : BD 0BAB G D :D:B A 2, :D EJ - 4D AE BAE BA 2B7B E B7 A : I 0 D : : D: 0BAB G D 0J

12. 1@2 A / 807 5 / o Xc 11 yf -D E D ep i rvep TZ hRu a mtV s S k nlgdwX 11 s P 11 0 P EL , 0EDE: 0 D 8 E " N D D CEDE: K E EC M L D , C D EC D , EL 7 K . A DE " 11 0 3 C D CEDE: 0 L D : ED 23

13. 1@2 A / 807 5 / • A f • a 1 • s S • M L M S A s l ! "#:%, '|)#:%, *#:% s V uf 3 il ! "%, '|)#:%, *#:% S a

14. 1@2 A / 807 5 /

15. 1 2 M @ P / 7 08 5 L / ( ( ) + FN D 1N 2 A

16. 1@2 A / 807 5 / , , 1 2 9 9 9 • I M AC ., ( 0 06 2 06 ))

17. 1@2 A / 807 5 / , 2 21 1 1 • F M C . - ) 0 0 1 0 ( 7 1

18. 1@2 A / 807 5 / .- - 8 • F P M CR St. Peter’s Basilica Trevi Fountain Colosseum Dubrovnik Piazza San Marco 0 ) 8 2 2 2 (. , 1

19. 1@2 A / 807 5 / 1 9 3 9 9

20. 1@2 A / 807 5 /

21. 1@2 A / 807 5 / 1 - 2

22. 1@2 A / 807 5 / 2 3 • ! " #$ %$ &$ '$ !(, "( * 3 3 + = - . / 1 +1 +2 +12 = +2+1 31 2 3 * = *4 = *5 3

23. 1@2 A / 807 5 / ! " # 1 = &' ( ") 0 &+ #) 0 0 1 ,-- ,-. ,-/ 0- ,.- ,.. ,./ 0. ,/- ,/. ,// 0/ 1 2 3 1 !4 = 56 = 7 8|: 6 6 = 1, 2, 3, 1 < 4 = ", #, 1 < 5 = 7 8|t 2 2 3 2

24. 1@2 A / 807 5 / - bf hIZ i Ii [ T or I g f] p M xq VaeP l eaI w • Pu 4C 2 4 4 4 A4 4 2 " , )(" 0 • 4 P 2 4 4 O !" !# mrP n $" $# t % [ cI ox &', &), &*, +', +)

25. 1@2 A / 807 5 / 5 3 2 • !" !# $# !" % % ? % ? !# $# '" '# () (*

26. 1@2 A / 807 5 / e ) ( • a ! M a 2 "|$ d • F • %& = ()* )* 6 F )& + %& = 0 -& .& 1 0** 0*& 0*1 0&* 0&& 0&1 01* 01& 011 -* .* 1 = )& +()* = 0

27. 1@2 A / 807 5 / n 2 )( ( ( • ! e • a F • aCl7 mM i VCO • ) 2 )( ( ( • d a p M n" p #$ % #% % #$ % &% % #$ % &$ % #% % &$ % &% % &$ % #% % &% % 1 ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ #$ ) #% ) #$ ) &% ) #$ ) &$ ) #% ) &$ ) &% ) &$ ) #% ) &% ) 1 *%% *%$ ⋮ *++ = "- = 0

28. 1@2 A / 807 5 / e i ( 3 4 5 ) 84 • V • e i ! s • 3 2( 3 4 5) n • R lidt c NS Ab • A p ES O s b • ! M a R "|$ c • A C S r A x ! = &' ( )&* +* +' +*+' +* +'′ +*+'′ 2 - - - -

29. 1@2 A / 807 5 /

30. 1@2 A / 807 5 / (30 - ) - (

31. 1@2 A / 807 5 / + 3 2 + 1 + D 1 + 3 2 - 1 + 1 + D

32. 1@2 A / 807 5 / - B R U 2 • i F 3 . , i 3 ,, B IE O T S IE O S

33. 1@2 A / 807 5 / 3 • 3! +2" • " - #, % D

34. 1@2 A / 807 5 / v c p ae atS A mSls • ( 4 - (3 ( 3 ( ( r • atE p t !b N mSls ( ( ( ( • t P • ( (E) Sn R Si • t S ( ( mSls • 3 ( ( C 3 ( () - o

35. 1@2 A / 807 5 / - a P • 3 5 • ) ( ) !" # !$ %" %$ &" −!" 0 &$ 0 −!$ # )" )$ = 0

36. 1@2 A / 807 5 / - • 6 3 6 3 ! "R$, "&$ "'( )$( ) *$, '( ! = , - ∑$ ∑( /$( 0/$( /$( = ) *$, '( − )$( '( *$

37. 1@2 A / 807 5 /

38. 1@2 A / 807 5 / 3 3 38 C A B 2 + C A+ B 8 + GF GF 2 + D A B / D A B + A B 2

39. 1@2 A / 807 5 / riMl v v SU3 /_ [ p mp • hMgeM P R ] u[ s 3 / P t • + + + . HH G H I9 : GC A 9IC A G & 0 3C A H A 3.,,2 6 Photo Tourism: Exploring Photo Collections in 3D Noah Snavely University of Washington Steven M. Seitz University of Washington Richard Szeliski Microsoft Research (a) (b) (c)brp rv S N cop N y npafrd

40. 1@2 A / 807 5 / ( ( ) - 4 Gd G B G • nfGr E • a a aS • S G • G S c • S o G uq t l G M • bi 4 4 0 g e

41. 1@2 A / 807 5 / ( f I M • P ) C) • 2 342 342 342 3 • S S1 / V1 / / • ) ) ) G1U • I1 G1M

42. 1@2 A / 807 5 /

43. 1@2 A / 807 5 / n V V • hp D P P C • oa VG ft PV S • / r O M P • 43 / e / • / 2 / + P

44. 1@2 A / 807 5 /

45. 1@2 A / 807 5 / W - 5 2 2 4 2 2 4 2 . 2 5 + -M TP 1 H S

46. 1@2 A / 807 5 /

47. 1@2 A / 807 5 / 1 _ V V 4 1 1 2 327 4 V S V 2 2 327 4M 2 2 2.7 4 OM S 7 S M OM

48. 1@2 A / 807 5 / n M L IS • • + 4 • O cf p • r oGe8 C • rG bGe8 a • m • G t • l

49. 1@2 A / 807 5 / L o RlDI f n V • V U • A 9 9 Rc G 3 P ge • A 9 CA 94 CA 9 9 CA 9 9 9 • MO Fmh • U O U a i 9 9 3 S S

50. 1@2 A / 807 5 /

51. 1@2 A / 807 5 / ls a D br g LI xTwo R • D U a • a Da • m gc • eh d gc • t iElh psk g nu gc • .5 . .1 .5 .5 : • - . .1. 5 / 5: 5 1 A • @5 .

SLAMの概要と画像を用いた3Dモデリングの基礎

Ken Sakurada

SLAMの概要と画像を用いた3Dモデリングの基礎