Face Detection And Tracking Method Research Based On Video Sequences

With the development of computer and information technology, and the needs of applications in the related area, face detection and tracking technology has attracted many researchers. Home and abroad has many different research ways about this technology, and the new research results are recurring. Human face provides broad application values in many research fields such as video surveillance, intelligent human-computer interaction and computer vision.While face detection and tracking has been achieved many results, there are several research difficult problems as following:real-time, robustness and occlusion handling. In this paper we have made a deep study on these issues, and proposed a set of face detection method based on a combination of mixed skin color models and of gradient features. Results show that the method improved the accuracy of detection.On using of the detection algorithm, we proposed and validated the face tracking method based on the improved Condensation filter. It is fully taken into account the non-rigid facial features and the complexity of the tracking environment.After drawing on some algorithm of the predecessors, we made some key improvements for the requirements and the experiment conditions of this paper. Major research works are as follows:1 In reference to a large number of face detection methods, for the real-time requirements of the video application background, we proposed and implemented the face detection algorithm based on the combination of color and gradient features. At first, we separate out the color-related areas using color features from the complex images. Secondly, we obtain the face region combined with the gradient information and the morphology ways. Experiments show that the algorithm has fast response, high accuracy, and it is suitable for human face processing system as a pre-processing and detection.2 In tracking way, firstly, we determine the tracking initial position using the face detection algorithm and the inter frame differentiation; then we apply the improved Condensation filtering algorithm to estimate the human face motion and achieve the face tracking of video based on the detection algorithm. This method has high accuracy, fast tracking speed, and overcome the effect of the human face rotation and part occlusion. Experiment show that this algorithm has better robustness and fast tracking speed compared with the tracking using skin color model as the single-track trail.