Robust and Real Time Face Tracking Using Particle Filter Based on Probablistic Face Model

This paper presents an algorithm for real time and robust human face tracking against pictures and other objects. It is based on Haar-like features, skin segmentation and motion information for face detection. Face tracking is performed using particle filter which depends on skin color and probabilistic face model. Basically, the employed features for face detection are Haar-like. We employ PCA to extract the most significant features. The extracted features are then used as learning vectors of neural networks. Neural networks classify the objects to face or non-face. Face detector locates faces from the face candidates determined in first frame by using skin color information. Then the detected face is tracked using particle filter and based on the probabilistic face model which is updated using the information related to variability with respect to head rotation, illumination, facial expression and occlusion. The proposed method is done in real time and achieves high performance. Also it is robust against illumination variations and geometric changes. Experiment results show that this algorithm is the best in comparison with other existing methods specially when there is occlusion. Also the robustness of the proposed method, when the tracked face is close to unicolor objects, is proved.