Face Tracking And Face Recognition In Monitoring System

With the rapid development of the information technology, face analysis hasbecome one of hot topics in the field of computer vision and pattern recognition. Ithas important research value in the fields such as personal identity, intelligentinteractive systems, video surveillance and so on. Face analysis contains manysubjects such as face detection, face tracking, face recognition animation, facialexpression analysis. This project comes from the video surveillance system by thepublic security department, the system is used to long-term monitor the peoplecoming in-and-out of some fixed place and it contains face detection, face tracking,face extraction and face recognition. For this project, some research is done in thispaper and summarized as below:1. Introduction of the face detection based on complex background. First, wediscussed several common color models and presented the basic process of facedetection, then analyzed the face detection based on skin color likelihood probabilityin detail.2. Introduced some common face tracking algorithm, and we focused on theanalysis of the principle and the steps of the Mean Shift algorithm, then theadvantages and the disadvantages of which are discussed.3. An improved algorithm of the Mean Shift was proposed which was based onthe tracking target real time updating and was combined with the LBP texturalcharacteristics. The tracking accuracy could be improved through updating thetracking target. Besides, the initial tracking target was divided into several pieces toenhance the target’s local features, it could improve the robustness of the tracking tothe skin-color like background. Then we analyzed the Mean Shift’s disadvantage thatthe algorithm could not recovery tracking rapidly after the loss of the tracking targetand applied the frame difference method after target loss to solve this problem.4. Introduced several face recognition algorithms based on subspace and animproved face recognition algorithm was proposed which was a2D-LDA algorithmbase on block and sample expansion to improve the face recognition accuracy indealing with single training sample, the local features could be enhanced throughdividing the face into several blocks.5. Designed the system framework and gave its software implementation andverified its feasibility.