This project has two components namely video-based face recognition and activity recognition. The aim of the first component is to exploit spatiotemporal information in surveillance videos for accurate face recognition. The second component aims at the classification of human activities in surveillance videos. Funding is available for a PhD student to work on the second component [details]. Please contact Dr. Ajmal Mian directly.

Face Tracking with a PTZ Camera

Surveillance cameras have a wide field of view and do not capture the human face with sufficient resolution. PTZ (pan, tilt and zoom) cameras can be used to track a human face and acquire frames at high resolution. This is challenging because with zooming, the field of view of the camera becomes very narrow and faces can be easily lost. Due to this difficulty existing literature contains PTZ tracking with mutliple cameras where additional cameras ensure that the location of the face is known with one camera zooms in on the face. We have proposed an algorithm that uses a single PTZ camera for face tracking. The algorithm is implemented in OpenCV and can track a face in real time and zoom in on it. Sample demos and source code can be downloaded.

Download demo videos. (6.6MB zip file)
Download C++ code. (requires OpenCV libraries and Sony SNC-RZ50P network camera)