Intelligent people surveillance framework using multiple cooperative cameras

An integrated framework for intelligent people surveillance using multiple cameras is proposed in this thesis. The cameras are geometrically related through joint efforts of multi-camera calibration, ground plane homography constraint, and field-of-view lines. As a result, they work in a cooperative manner under passive scene view mode or active object view mode. Under the passive mode, object detection is implemented using an adaptive Gaussian mixture model. Based on a study of multi-view object association, the detected objects are consistently labeled and projected onto a virtual top-view of ground plane. Under the active mode, a color-based particle filtering algorithm is utilized to constantly track an object. A novel relay strategy is developed by combining the tracking algorithm and the pan-tilt-zoom capabilities, and therefore the object being tracked can be handed over between cameras. Finally, a preliminary study is also carried out on anomaly detection and localization via wearable wireless motion sensors.