Fast Facial Feature Tracking with Multi-Cue Particle Filter

The paper represents an effective and robust facial feature tracking approach based on the multi-cue particle filter. Both color and edge distributions are integrated into the filter to ensure the tracking accuracy. Sub-region color model is used to rapidly depict the spatial layout of each facial feature. Furthermore, the paper uses edge orientation histogram as a complementary feature to enhance the robustness of tracking results. In order to enhance the robustness of our work, we propose a point distribution model to constraint face configuration and avoid tracking fails during occlusion. An efficient updating algorithm is introduced to avoid tracking error accumulating problems. Experiments show that the method based on the multi-cue particle filter and the updating algorithm gives us an inspiring tracking result. Compared with other facial feature tracking approaches, the method has the good performance in long-time facial feature tracking with temporary occlusions