关键词: 递进结构, 均值权重, 跟踪校正, Gabor特征, 隐马尔可夫模型（HMM）

Abstract: Based on video surveillance, this paper designs a kind of face detection, tracking recognition system, the function of which is detecting and tracking face image real-timely in video, at the same time for recognizing for the identity. The cascade classifier that Gentle AdaBoost algorithm constructs has a problem of low efficiency. To deal with that it proposes a cascade classifier in progressive complexity. In view of the highest weight particles of the traditional particle filter may not be accurate, the weight equal value of particle filter is proposed. Aiming at the sample recession problem of traditional particle filter, it puts forward a strategy of tracking correction that combines face detection and face tracking algorithm. The method based on Gabor transform and the HMM is used to identify the faces which are detected and tracked. The experiment results show that in the video, the system is capable of detecting the faces accurately and tracking them real-timely, it can also realize the fast recognition. It is an effective method that can be applied to the video surveillance system.

Key words: progressive structure, weight equal value, tracking correction, Gabor feature, Hidden Markov Model（HMM）