2011 3 A Graduation Thesis of College of Engineering, Chubu University Pose Estimation by Regression Analysis with Depth Information Yoshiki Agata
CG [2] [3][4] 3 3 [1] HOG HOG TOF(Time Of Flight) iii
1 TOF 2 3 4 iv
1 TOF 1 1.1 TOF SDK............................... 4 1.2........................ 4 1.3............................. 6 1.4........................... 7 2 9 2.1................................ 9 2.2........................... 10 2.3................... 10 2.3.1 Histograms of oriented gradients(hog)............ 10 2.3.2........................ 12 3 13 3.1 3................................ 14 3.1.1 3........................ 14 3.1.2 Poser............................ 14 3.1.3 opengl.................... 15 3.1.4 3............. 18 3.2................................. 21 3.2.1................... 22 3.3............. 23 4 25 4.1..................................... 25 v
4.2......................... 26 4.2.1..................... 27 4.3.............................. 29 4.3.1................................. 29 4.3.2................................. 29 4.3.3.............................. 30 35 37 39 vi
1.1 TOF (SR-4000)......................... 2 1.2 TOF.............................. 3 1.3 V=1.0 HSV.......................... 5 1.4 HSV............................ 5 1.5.................. 6 1.6............................. 7 1.7..................................... 7 2.1 HOG................................... 11 2.2............................ 12 3.1................................. 13 3.2................................... 14 3.3 Poser............................... 15 3.4.............................. 16 3.5...................... 17 3.6........... 17 3.7 3.......................... 18 3.8 3.............................. 19 3.9............................. 20 3.10................................... 20 3.11...................... 21 3.12................................. 21 3.13................................. 22 vii
3.14............................ 22 4.1..................................... 26 4.2........................... 27 4.3............................ 28 4.4.............................. 29 4.5.................. 30 4.6.............................. 31 4.7............................ 32 4.8.............................. 33 viii
4.1 Recall................................. 30 ix
1 TOF TOF TOF LED 1.1 mesa Swiss Ranger SR-4000 1.2 TOF 10MHz 1 LED TOF 1
1 TOF 1.1: TOF (SR-4000) 2
1.2: TOF 3
1 TOF 1.1 TOF SDK TOF TOF SDK(Software Development Kit) SDK TOF SDK API(Application Program Interface) SDK API typedef struct CMesaDevice SRCAM SR-4000 int SR OpenUSB( SRCAM srcam, unsigned int serialnumber ) SR-4000 int SR Close( SRCAM srcam ) SR-4000 int SR Acquire( SRCAM srcam ) unsigned int SR GetCols(SRCAM srcam) unsigned int SR GetRows(SRCAM srcam) ( ) int SR CoordTrfFlt( SRCAM srcam, float* x, float* y, float* z, int pitchx, int pitchy, int pitchz ) (x y z ) 1.2 TOF 0m 5.0m (z ) 0 255 HSV 1.3 HSV 1.4 4
1.2. 1.3: V=1.0 HSV 1.4: HSV 5
1 TOF 1.3 TOF SR-4000 SR-4000 1.5[1] TOF 1.5[2] 2 SR-4000 1.5: 6
1.4. 1.6: 1.4 TOF 5m TOF 4 5m 1.7 1.7: 7
2 2.1 3 1 2 3 9
2 2.2 [1] [5] HOG HOG HOG 3 HOG TOF 2.3 HOG [7] HOG 2.3.1 Histograms of oriented gradients(hog) [1] HOG HOG [6] ( 2.1 (b)) HOG I(u, v) m(u, v) θ(u, v) (2.1) (2.2) 64 128 10
2.3. m(u, v) = I u (u, v) 2 + I v (u, v) 2 (2.1) θ(u, v) = tan 1 I v(u, v) I u (u, v) I u (u, v) = I(u + 1, v) I(u 1, v) I v (u, v) = I(u, v + 1) I(u, v 1) (2.2) (2.3) 2.1: HOG m θ (2.4) c (p p ) θ v c (θ ) = u m(u, v)δ[θ, θ(u, v)] (2.4) v δ Kronecker θ θ 1 (2.4) θ N V c = {v c (1), v c (2),, v c (N)} ( 2.1 (d)) θ 0 360 0 180 c 8 8 θ 8 (2.5) (q q ) 2 2 v c(n) = v c (n) q q N v c (k) 2 + ɛ k=1 (ɛ = 1) (2.5) 11
2 1 V c V c = {v c(1), v c(2),, v c(b N)} B 2.3.2 [7] 2.2 2 2 2 (2.6) Bhattacharyya m S = pu q u (2.6) u=1 16 16 2.2: 12
3 3 2 TOF 3.1 3.1: 13
3 3.1 3 3 3 3.1.1 3 3 3D Poser 3 3.2 3 3.2: 3.1.2 Poser 3D Poser Poser 3.3 3.4 14
3.1. 3 3.3: Poser Poser 3.1.3 opengl Poser opengl α aspect x c y c z c φ x φ y φ z h m θ x w y w z w T bg T in 15
3 3.4: z c 3.5 3.6 16
3.1. 3 3.5: 3.6: 17
3 3.1.4 3 3 3 depth map 3.7 3 19 1 (x, y, z) 3 1 19 3 = 57 3.7 3 3.8 TOF 3.9 3.10 3.7: 3 18
3.1. 3 3.8: 3 19
3 3.9: 3.10: 20
3.2. 3.2 3.12 2 32 32 C 2 496 3.11 (a) (b) 3.11: 3.12: 21
3 3.2.1 16 16[pixel] 2 3.13 3.14 3.13: 3.14: 22
3.3. 3.3 496 x = (x 1, x 2,..., x 496 ) 57 y = (y 1, y 2,..., y 57 ) n 496 n X = (x 1, x 2,..., x n) T 57 n Y = (y 1, y 2,..., y n) T X Y (3.1) 496 57 A A = arg min A AX Y 2 (3.1) A TOF X Y Y (3.2) ε Y = A X + ε (3.2) 23
4 4.1 Recall( ) 1 0 true positive total pixel 4.1 A B true positive total pixel 4.1 Recall Recall 1 Recall = true positive total pixel (4.1) 25
4 4.1: 4.2 10[ ] HOG 4.2 26
4.2. 4.2: 4.2.1 10 100[ ] 10[ ] 27
4 HOG 4.3 Recall 60[ ] 4.3: 28
4.3. 4.3 4.3.1 100 60 HOG (D-HOG) HOG (HOG) (DDF) 3 4.3.2 4.4 4.5 Recall Recall Recall 4.1 4.4: 29
4 4.5: 4.1: Recall DDF D-HOG HOG WAVE 0.76 0.67 0.63 WALK 0.70 0.60 0.53 4.3.3 4.6 HOG D-HOG 4.7 4.8 30
4.3. 4.6: 31
4 4.7: 32
4.3. 4.8: 33
recall HOG 0.13 D-HOG 0.09 HOG 0.17 D-HOG 0.1 35
37
[1],,,, HOG 3,MIRU2008,pp.960-965,2008. [2],,,,, MIRU2006, pp.70-77, 2006. [3],, tree-based fil- tering, MIRU2006, pp.63-69,2006. [4] M. Andriluka, S. Roth, and B. Schiele, Pictorial structures revisited: People detection and artic- ulated pose estimation CVPR,pp.1014-1021,2009. [5],,,, 3 MIRU2010 pp.589-594 2010-07 [6] N. Dalal and B, Triggs, Histograms of Oriented Gradients for Human Detection, IEEE Computer Vision and Pattern Recognition, pp. 886-893, 2005. [7],,,SSII2009,,2009. 39
( ) 2011 3