Tracking all traffic: computer vision algorithms for monitoring vehicles, individuals, and crowds

This article presents a vision-based system for monitoring crowded urban scenes. The approach combines an effective detection scheme based on optical flow and background removal that can locate vehicles, individual pedestrians, and crowds. The detection phase is followed by the tracking phase that tracks all the detected entities. Traffic objects are not simply tracked but a wealth of information (position, velocity, acceleration/deceleration, bounding rectangle, and shape features) is gathered about them also. Potential applications of the methods include intersection control, traffic data collection, and even crowd control after athletic events. Extensive experimental results for a variety of weather conditions are presented. Future work would focus on methods to deal with shadows and occlusions.