Research On Algorithms For Facial Feature Point Tracking Under Complex Conditions

With the rapid development of information technology and artificial intelligence, computer has the ability to simulate human thinking and wisdom. Computer vision has become one of hot research topics. In recent years, face detection, face recognition, feature extraction, and facial feature tracking methods are becoming much richer. Facial feature point tracking is a basis for the subsequent face recognition and facial expression analysis. Researchers have proposed many tracking algorithms. However, in real applications, because of the variation of the face posture, occlusion, facial expressions and other internal and external factors in the video, it makes the facial feature point tracking become more complex Thus, the issue of facial feature point tracking under the complex background deserves more research than it has received.The research work and innovation of this paper are mainly reflected in the following aspects:1. A comparison of main facial feature point tracking algorithms is presented. The particle filter algorithm is compared with the EKF, UKF algorithms on the tracking error rate by simulating experiments. By a comparison of Meanshift, optical flow and KLT methods, it is shown that the particle filter selected as the face feature point tracking method is effective.2. A particle filter facial feature point tracking algorithm is proposed under color and texture with a shape constraint. Because the nose feature points change very little in the whole tracking process, a shape constraint model is built for the eyebrows, eyes, mouth, and other key feature points by using them as reference points. Then the color and texture characteristics of each region of the feature points are extracted. The similarity for each feature point between the previous frame and the next frame is calculated. The position with the highest similarity is selected as the optimum target position of the feature point. When the error of the feature point exeeds a certain threshold of constraint conditions in the shape constraint model, the ASM algorithm is restarted and then the particle filter tracking is continued. Experimental results show that the algorithm has good accuracy and robustness and can effectively relieve the impact of changing complex background with changing posture, facial expression and occlusion.3. An improved condensation algorithm for facial feature point tracking is presented. The incremental PCA algorithm is employed to update the mean and eigenbasis of the tracking feature points online. The depth and breadth of feature points in the subspace are extracted as the observed values. A forgetting factor is added as the basis for the balance of old samples and new samples. Experimental results show that the algorithm has good robustness under the complex background with changing posture, facial expression and occlusion.