A novel human face tracking system with multiple cameras is proposed in this article. Unlike most surveillance approaches requiring overlapping field of views (FOVs) for tracking targets across multiple cameras, our proposed system utilizes face recognition as cue to build correspondence between cameras, and is therefore compatible with application with non-overlapping FOVs. Face recognition achieves great success in recent years, however it has an intrinsic limitation of low tolerance to face pose changes. In the proposed system, face reconstruction technology is adopted to transform non-frontal human face to a frontal one to substantially enhance the usability of face recognition, so that correspondence of humans across multiple trajectories can be reliably established.
展开▼