Introduction


Project Name :	Using Face Recognition Result as Cue for Human Tracking System with Multiple Uncalibrated Cameras (ITS/497/09)


Objective :	The objective of this project is to tracking human across multiple cameras in a wide area. Surveillance of wide area requires a network of cameras, which are not always having overlapping camera views in practical environments. Tracking problem can be simplified if cameras are properly calibrated together with an accurate scene model, but since the cameras could be adjusted or moved due to PTZ actions, rendering manual calibration after each camera movement to be almost infeasible in most of the situations. Thus, it is more preferable that the tracking system can self-calibrate cameras and complete site modeling automatically to adapt to any camera or scene change due to normal surveillance operations. On the other hand, existing systems usually employ the appearance of persons as cue to establish correspondence between cameras. However, appearance of different persons could be similar in practical scenarios, which will cause ambiguity and error in tracking. This project aims at resolving this problematic issue by using human face recognition result as a strong cue for correspondence establishment between cameras. Specifically, we shall employ face reconstruction technology to transform human face from arbitrary pose back to a frontal one to substantially enhance the usability of face recognition to establish strong correspondence of humans across multiple uncalibrated cameras. We expect the proposed tracking system is more reliable than existing methods.

Project Duration :	12 months