Video Image Facial Feature Point Tracking Technology Research

Face detection, tracking and facial feature point locating has become one of key technology of the current computer vision and has been widely used in many research works , such as intelligent surveillance system, biometric recognition technology, expression analysis .While most existing algorithms usually fail in the presence of cluttered background, adverse circumstances and part of shading. We aim at resolving the current problem of face detection, tracking and facial feature point locating under complex observation conditions. Some methods are proposed to improving these algorithms.Complex illumination environment is a main factor of face detection. In order to solve this problem , this paper use mulit-Retinex image enhancement algorithms. Face detection is a non-human face classification and face the problem and need to detect the faces real-time in the image The Adaboost algorithm can detect faces very accurate. However, the Adaboost algorithm become very slow because the classifier need tremendous computations, and it is not a real time algorithm. In order to improve the detection speed and accuracy of Adaboost, skin color segmentation used in this paper in the process of face detection. First the skin area can be division in a picture, and then the Adaboost algorithm will be used in this area. Experimental results show this improved Adaboost algorithm make the progress faster and increases the face detection rate .Optical flow algorithm has long been used in object tracking . There is a natural tradeoff between local accuracy and robustness when choosing the integration window size. The feature point may drift and be declared"lost". This paper uses a pyramidal implementation of the classical optical flow algorithm. This algorithm can improve the accuracy of facial feature trackingIn a video sequence, the posterior density of object tracking is usually non-Gaussian and non-linear. Particle filter can model accurately in non-linear non-Gaussian system, and be widely used in object tracking. However , in the tracking progress, the object's appearance can change such as the presence of significant variation or surrounding illumination. Fixed appearance models of the target trained cannot fix the object's appearance. In this paper, we used incremental PCA efficiently adapting online to changes in the appearance of the target. This method can significantly improve the accuracy and robustness of face tracking. The active shape model (ASM) has been widely used to track a face from a video sequence. However, it is usually limited to frontal view or the cases of small-scale head movement。ASM may lose in condition of fast motion and significant variation of the object's appearance . Based on the improved face detection algorithm and the particle filtering tracking of incremental PCA algorithm , We propose an enhanced ASM to meet those challenges. Extensive experiments demonstrate the flexibility and accuracy of the proposed method。...