社団法人人工知能学会 Japanese Society for Artificial Intelligence 人工知能学会研究会資料 JSAI Technical Report SIG-Challenge-B001-6 (5/3) 3 RoboCupSoccer SSL Humanoid 3 A Realtime Acquisition of 3D Shape of Humanoid Robots Based on Multiple External Camera Images Aiming at 3D Information Server for the RoboCupSoccer SSL Humanoid Takaumi KIMURA Yasuhiro MASTANI Osaka Electro-Communication University Abstract RoboCupSoccer SSL Humanoid, soccer game by humanoid robots using external cameras, was proposed. Although 2D image processing are used in the current phase, in the next phase, the targets are real-time 3D shapes acquisition based on images of multiple cameras and deciding action based on the results. A shared vision system is also the target. However, what to be shared for 3D vision system has not been decided. In this paper, the authors propose that voxel data of 3D space on the game field is shared. Moreover, in order to prove the realization of the proposal, the authors develop a program for cutting voxels whose size is 10[mm] of 3[m] 4[m] 0.5[m] space based on images taken with 4 cameras by visual cone intersection and sending run-length compressed data to clients via network. This program can execute the chain of processing 30times per second on PC with Intel Core i7 950 processor. 1 RoboCupSoccer SSL: Small Size robot League SSL SSL Humanoid [, 2010] SSL 2 SSL Humanoid 2015 3 SSL 2010 SSL-Vision SSL-Vision [Zickler, 2010] SSL Humanoid 2 SSL-Vision 3 3 3 2 2009 SSL Humanoid Figure 1 2 33
RoboCupSoccer 3D Figure 1: Initial phase of the SSL Humanoid (2009) 3 2.3 Figure 2: Final phase of the SSL Humanoid (2015) 2015 SSL Humanoid Figure 2 3 2.1 2.2 3 3 3 2 3[m] 4[m] 0.5[m] 400[mm] 1 10[mm] 1 30 34
3 3 3.1 3 [, 2006][, 2001] Volume Intersection Method(VIM) Space Carving Method(SCM) 2 SCM ( ) SCM 3.2 Z.Zhang [Zhang, 2000] (c x,c y )[pixel] (f x,f y )[pixel] (k 1,k 2 ), (p 1,p 2 ) [R t] 3.3 3.2 c, (x, y, z)[mm] (i, j, k) s[mm] (1) x = is, y = js, z = ks (1) (x, y, z) (X, Y, Z) (2) (9) c X x Y = R y + t (2) Z (u, v) (3) (9) X = X/Z (3) Y = Y/Z (4) R = X 2 + Y 2 (5) z X = X (1 + k 1 R 2 + k 2 R 4 ) +2p 1 X Y + p 2 (R 2 +2X 2 ) (6) Y = Y (1 + k 1 R 2 + k 2 R 4 ) +p 1 (R 2 +2y 2 )+2p2 X Y (7) u = f x X + c x (8) v = f y Y + c y (9) table c (i, j, k) =(u, v) (10) 3.4 2 (R cuv,g cuv,b cuv ), (R cuv,g cuv,b cuv) D th S cuv (11),(12) D cuv = R cuv R cuv + G cuv G cuv S cuv = { 3.5 + B cuv B cuv (11) 0 (D cuv >D th ) 1 (D cuv D th ) (12) V ijk {0, 1 1 1 0 C for(i, j, k){ V ijk =1; for(c){ for(i, j, k){ if(v ijk == 1){ (u, v) =table c (i, j, k); if(s cuv == 1){ V ijk =0; 3.6 35
Figure 3: Experimental enviroment Table 1: Specifications of experimental cameras Company Creative Product name Live!Cam NoteBook Ultra(VF0490) Interface USB2.0/1.1 Resolution 640 480 1 1[byte] 1[bit] 7[bit] 128 127 3.7 UDP 1 s[mm] 3 [mm], [byte] 4 SSL Humanoid 4.1 SSL Humanoid 3025[mm] 4050[mm] 500[mm] Figure 3 4 Table 1 PC PC Table 2 2 PC PC DirectShow USB 4 Figure 4: Timing chart PC 3 4.2 1 1 55 60[ms] PC CPU 4 HTT 8 8 Figure 4 4.3 640 480[pixel] 1 10[mm] 3 1 ex1 0 ex2 6 6 ex3 10 6 4 378[mm] MANOI AT01 360mm KTR-1HV 36
Table 2: Specifications of experimental PCs Server PC Client PC Operating System Windows 7 Professional 64bit Windows Vista Home Premium 32bit SP2 Main Memory 3GByte 3GByte Graphic Board NVIDIA GeForce GTX260 Mobile Intel 4 Series Express Chipset DirectX DirectX 11 DirectX 11 CPU Intel Core i7 950(3.19GHz) Intel Core Duo U9400(1.4GHz) Figure 5: Image of camera 1 (ex3) Figure 9: Result of voxel cut from lower view point (ex3) Figure 6: Image of camera 2 (ex3) Figure 10: Result of voxel cut from higher view point (ex3) Figure 7: Image of camera 3 (ex3) 9 Figure 10 1 Figure 11 ex1, ex2, ex3 48526[byte], 57929[byte], 62374[byte] Figure 12 Figure 13 4.5 Figure 8: Image of camera 4 (ex3) 4.4 ex3 Figure 5 8 Figure Figure 11 SSL Humanoid 2010 6 10 1 30 730[Kbyte] 37
time[ms] 45 40 35 30 25 20 15 10 5 ex1 ex2 ex3 0 0 100 200 300 400 500 600 700 800 900 1000 cycle Figure 11: Processing time Figure 12: Voxels of humanoid robot at the center 5 RoboCup SSL Humanoid 3 3 Intel Core i7 950 PC SSL Humanoid 1 10[mm] 1 30 60[Kbyte] PC [, 2010] : SSLH: RoboCup Soccer SSL Humanoid,, Vol.25, No.2, pp.213 219, 2010. [Zickler, 2010] S.Zickler, T. Laue, O. Birbach, M. Wongphati, and M. Velosso: SSL-Vision: he Shared Vision System for the RoboCup Small Size League, RoboCup 2009, Robot Soccer World Cup XIII, Springer-Verlag,2010. Figure 13: Voxels of humanoid robot at the corner 1/10 1 30 14[Mbps] Figure 12 1 10[mm] 400[mm] Figure 13 [, 2006] : :,, 2006. [, 2001],, : :,, Vol.42, No.SIG6(CVIM2), 2001. [Zhang, 2000] Z.Zhang: Calibration: A flexible new technique for camera calibration, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol.22, No.11, pp.1330-1334, 2000. 38