Detecting Pedestrians

Viola, Jones, and Snow recently implemented a pedestrian detection system that incorporates both appearance and motion in real-time. Simple sum-of-pixel filters are boosted into a robust pedestrian classifier. Detection is then achieved by thesholding a linear combination of these simple filters. The simplicity of the filters, along with some implementation tricks, enables the system to run in real-time. Motion information is incorporated by taking differences between successive frames in time. This paper is a reimplementation of their system, with the purpose of evaluating the merits and pitfalls of their approach. I also discuss some issues that were inadequately explained in their paper, such as how to train features used in the cascade, and provide a performance comparison.

[1]  Paul A. Viola,et al.  Rapid object detection using a boosted cascade of simple features , 2001, Proceedings of the 2001 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. CVPR 2001.

[2]  Xavier Binefa,et al.  Robust Real-Time Periodic Motion Detection, Analysis, and Applications , 2000, IEEE Trans. Pattern Anal. Mach. Intell..

[3]  Paul A. Viola,et al.  Detecting Pedestrians Using Patterns of Motion and Appearance , 2003, Proceedings Ninth IEEE International Conference on Computer Vision.