Face Recognition And Tracking Based On Video

With the rapid development of information technology, video surveillance is more and more used in security surveillance and identifying related fields, and face recognition is one of the important biological feature recognition method, mainly on the basis of people’s facial features, the automated method to verification or identification of a person, with convenient using, the recognition rate and high non contact characteristics and broad development and application prospect. Most of the face recognition research work is carried out under good lighting conditions and positive attitude. However, in practical application in the real scene and not every time we can obtain reliable facial appearance description. Face recognition is one of the most important task is to find effective and distinctive facial appearance descriptions, which can reduce the low quality of the face image, scale change, light illumination changes, pose variations and the impact of partial occlusion. It has two main methods: based on the geometric feature description and based on the appearance of the shape description. In the process of facial changes geometry description method what has difficult in extracting reliable feature information, and the emergence of Eigenface due to residual spatial registration error, often bluring the image details, but single feature and can not effectively describe the facial information, and has low accuracy and poor robustness characteristics. This paper is mainly divided into two modules, including face detection and trackings, face feature extraction and recognition, as shown below:1) Face detecting and tracking modules: It is good security and surveillance application based smart module. This paper deeply analyzes the theory of Adaboost algorithm and its derivation of, and then selected the video analysis of the experimental results, which proved the advantages of good real-time performance. Face tracking is an important part in the intelligent video surveillance system, firstly we were in depth analysis of the basic theory of MeanShift algorithm, and then extended to the Camshift algorithm. Although traditional meanshift algorithm through manual settings object tracking, you could track the trajectories of moving objects, but it was not able to easily keep track of objects, such as object occluded or the same color or interference of obstacles. According to the characteristics of AdaBoost and Camshift algorithm, this can accurately track the automatic detection of face.2) Face feature extraction and recognition module: In order to solve the single feature based on video face recognition’s low accuracy, this paper proposed based on Local ternary pattern(LTP) features and Histogram of oriented gradient(HOG) features fusion method for face recognition. Combined with the contour and texture information, the HOG and LTP adaptive feature fusion method not only can accurate face recognition, solve the problem of low recognition rate, but also can reduce the training time, improve the speed of the system.In order to illustrate the effectiveness of the method in this paper, we respectively on Yale and ORL data set for testing, these data sets contain a lot of lighting changes, similar to those taken under uncontrolled conditions of natural images and from the Internet in a selected part of the video for simulation analysis. The experimental results show that: this method can improve the accuracy and reduce the error rate.