CVGIP2013 - A HEAD DETECTION SCHEME IN 3D ENVELOPE FROM DEPTH SIGNAL

at National Ilan University, Taiwan

Transcript

1. A Head Detection Scheme in 3D Envelope from Depth Signal Wei-Chi Lin(林暐智), Shih-Wei Sun(孫士韋), and Wen-Huang Cheng(鄭文皇)   Department of New Media Art and  Center for Art and Technology,  Taipei National University of the Arts, Taipei, Taiwan 國立臺北藝術大學 新媒體藝術系碩士班, 藝術與科技中心 Research Center for Information Technology Innovation, Academia Sinica, Taipei, Taiwan 中央研究院 資訊科技創新研究中心 a,b a,b c a b c
2. Outline •  Introduction •  Interactive Display System Based on Multi-Audience Tracking •  3D Envelope Detection from Depth Signal •  Experimental Results •  Conclusions and Future Work
3. Interac,ve  Display  System  Based  on     Mul,-­‐Audience  Tracking Proposed  System Conven,onal  System Too  low Too  high
4. Mo,va,on •  Interactive electronic bulletin •  head tracking scheme •  Related work •  People detection/tracking •  Video surveillance •  RGB-D camera Interac,ve  electronic  bulle,n
5. •  Motion-based method (Bayesian) •  Brostow and Cipolla [2] •  Optical flow of SIFT features for body detection •  Ozturk et  al.  [3] •  Color-based player tracking •  Santiago et  al.  [4] •  Homography-based people tracking •  Eshel and Moses [5] Video  Surveillance  Systems [2] G.J. Brostow and R. Cipolla, “Unsupervised Bayesian detection of independent motion in crowds,” IEEE CVPR, 2006 [3] O. Ozturk, T. Yamasaki, and K. Aizawa, “Tracking of humans and estimation of body/head orientation from top view single camera for visual focus of attention analysis,” IEEE ICCV, 2009. [4] C.B. Santiago, A. Sousa, L.P. Reis, and M.L. Estriga, “Real time colour based player tracking in indoor sports,” Computational Vision and Medical Image Processing, 2011. [5] R. Eshel and Y. Moses, “Homography based multiple camera detection and tracking of people in a dense crowd,” IEEE CVPR, 2008.
6. •  Pose recognition from depth images •  Shotton et al. [6] •  Kinect official sdk: Microsoft •  Depth image identification and localization •  Plagemann et  al.  [7] •  Kalman filter, hand tracking •  Park et al. [9] •  Depth-image frame differencing, object detection •  Baum et al. [10] RGB-­‐D  camera [6] J. Shotton, A. Fitzgibbon, M. Cook, T. Sharp, M. Finocchio, R. Moore, A. Kipman, and A. Blake, “Real-time human pose recognition in parts from single depth images,” IEEE CVPR, 2011. [7] C. Plagemann, V. Ganapathi, D. Koller, and S. Thrun, “Real-time identification and localization of body parts from depth images,” IEEE ICRA, 2010 [9] S. Park, S. Yu, J. Kim, S. Kim, and S. Lee, “3d hand tracking using Kalman filter in depth space,” EURASIP JASP, 2012. [10]  M.  Baum,  F.  Faion,  and  U.D.  Hanebeck,  “Tracking  ground  moving  extended  objects  using  rgbd   data,”  IEEE  MFI,  2012.
7. side view bird’s eye view Candidate  Camera  SeSngs
8. •  Occlusion handling •  frame-level •  Height variety handling •  Human subjects with different heights •  Still be detected •  Low complexity •  Extended to real-time applications Contribu,ons
9. 3D  Envelope  Detec,on  from   Depth  Signal  (1/3) color image depth image 3d model envelope detection
10. 3D  Envelope  Detec,on  from   Depth  Signal  (2/3) Head likelihood Depth Scanning 3D Representation of Depth
11. 3D  Envelope  Detec,on  from   Depth  Signal  (3/3) 3D Scanning
12. Experimental  Results •  Environment setting •  Scenarios: •  1-to-4-people •  347 frames •  8-people •  100 frames •  Kid-and-Holding-Baby •  30 frames
13. Qualita,ve  Results   Color channel Depth channel and detection result Head likelihood
14. Quan,ta,ve  Results  (1/2)  
15. Quan,ta,ve  Results  (2/2)   -­‐  [12]  C.  Wren,  A.  Azarbayejani,  T.  Darell,  A.  Pentland,  “Pfinder:  Real-­‐Time  Tracking  of  Human  Body”,          IEEE  T-­‐PAMI,  1997.   -­‐  [13]  I.  Haritaoglu,  D.  Harwood,  L.  Davis,“Who,  When,  Where,  What:  A  Real-­‐  Time  System  for  Detec,ng  and          Tracking  People”,  IEEE  FGRI,  1998.   -­‐  [10]  M.  Baum,  F.  Faion,  and  U.D.  Hanebeck,  “Tracking  ground  moving  extended  objects  using  rgbd  data,”  IEEE          MFI,  2012.  
16. Demo
17. Conclusions  and  Future  Work •  Higher detection rate •  Real-time •  Outstanding people detection (shoulder-by-shoulder) •  Interactive display system