Real-time face tracking under partial occlusion and illumination change

Abstract

In this paper, we present an approach which tracks human faces robustly in real-time applications by taking advantage of both region matching and active contour model. Region matching with motion prediction robustly locates the approximate position of the target, then active contour model detects the local variation of the target’s boundary which is insensitive to illumination changes, and results from active contour model guides updating the template for successive tracking. In this case, the system can tolerate changes in both pose and illumination. To reduce the influence of local error due to partial occlusion and weak edge strength, we use a priori knowledge of head shape to reinitialize the curve of the object every a few frames. To realize real-time tracking, we adopt region matching with adaptively matching density and modify greedy algorithm to be more effective in its implementation. The proposed technique is applied to track the head of the person who is doing Taiji exercise in live video sequences. The system demonstrates promising performance, and the tracking time per frame is about 40ms on Pentium ! 400MHZ PC.