Video Detection, Tracking and Classification

"In order to prevent, deter, and analyze suspicious behavior by individuals in public spaces, this project will develop a commercially viable system for tracking people in various environments using a single camera, or multiple cameras with overlappingfields of view. Single camera tracking of individuals requires algorithms for detecting people and then tracking them. The detection approach will make use of background scene models, skin tone detection, consistency of appearance, image motion and othervideo features. The tracking algorithm, building on the recent advances in particle filter tracking, will use a generic framework for simultaneous human tracking and verification. In order to track individuals using multiple cameras, the cameras must becalibrated relative to each other. In order to achieve this in a relatively automated, commercially-viable manner, corresponding point of a single moving object in each camera will be used to compute the relative calibration. In order to facilitatetracking of individuals through brief occlusions, and to identify tracked individuals as someone previously tracked, face verification will be incorporated into multiple-camera tracking. Two types of approaches will be used for face verfication: PrincipalComponents Analysis, a template-based approach, and Elastic Graph Matching, a feature-based approach."