Model Based Tracking

This paper presents an iterative method for tracking and classifying human activities in a video sequence. The basic idea is that activities can be positively identified from a sparsely sampled sequence of a few body poses acquired from videos. Connected Graph representation has been used to store the 2D human poses. These samples are matched against the graph abstractions derived from the frame where motion is identified in the video sequence. Sum of Absolute Differences method is used for motion detection in video frames. The probability of false activity detection drops exponentially with the increased number of sampled body poses. The proposed method gives very good results for activity detection in the surveillance video.

[1]  Paolo Remagnino,et al.  A multi-agent framework for visual surveillance , 1999, Proceedings 10th International Conference on Image Analysis and Processing.

[2]  Dorin Comaniciu,et al.  Real-time tracking of non-rigid objects using mean shift , 2000, Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No.PR00662).

[3]  Shyamsundar Rajaram,et al.  Human Activity Recognition Using Multidimensional Indexing , 2002, IEEE Trans. Pattern Anal. Mach. Intell..

[4]  Tomaso A. Poggio,et al.  Trainable pedestrian detection , 1999, Proceedings 1999 International Conference on Image Processing (Cat. 99CH36348).

[5]  Kaleem Siddiqi,et al.  Robust and efficient skeletal graphs , 2000, Proceedings IEEE Conference on Computer Vision and Pattern Recognition. CVPR 2000 (Cat. No.PR00662).

[6]  Dariu Gavrila,et al.  Pedestrian Detection from a Moving Vehicle , 2000, ECCV.

[7]  Michael Isard,et al.  Object localization by Bayesian correlation , 1999, Proceedings of the Seventh IEEE International Conference on Computer Vision.