Face Recognition In Intelligent Video Surveillance

Intelligent video surveillance is a hot research and application direction of computer vision which developed in recent years. Face recognition is a computer technology that through an analysis on visual features of human faces to complete identity authentication. The face recognition technology belongs to biometric identification and also is a hot research in recent years. The paper "face recognition in intelligent video surveillance" means to apply face recognition technology in intelligent video surveillance system and achieve real-time and automatic face recognition system. To apply face recognition technology in intelligent video surveillance system is not an easy task because it’s affect by illumination, noise and human face poses. Solving the problems brings to face recognition by intelligent video surveillance has great application value, and also is the research focus of this paper.This paper focus on face recognition in the special circumstances in intelligent video surveillance, the main contents are as follows:Firstly, after an analysis on the quality of images in intelligent video surveillance system and considering the algorithm’s effect and complexity, these follow pretreatment methods are used in this paper:gray histogram equalization was used to do gray level adjustment, median filtering was used to do noise suppression, and nearest neighbor interpolation and bilinear interpolation are used to do face size normalization. Experiments prove that the pretreatment methods are effectively, and will make face recognition easier.Secondly, to meet the requirement of the embedded application platform that the complexity of the algorithm must be low, a new algorithm based on statistical theory and LBP called SLLBP which is short for statistical liner local binary pattern was proposed in this paper after an analysis on the traditional LBP algorithm. Experiments show that the recognition accuracy and the recognition speed of SLLBP can meet the requirement of embedded systems.Thirdly, after a research focus on the influence brings to face recognition accuracy by face size scaling, a way of establish multi-level face database according to the face size was proposed in this paper, through the analysis on the influence brings to face recognition accuracy by different face poses, a method that collect as many representative face samples as possible during the process of face sample collection was proposed in this paper, and a way of using the image sequence information to solve a part of the problem of face false detection and sample incomplete was proposed in this paper, experiments show that these measures can improve the recognition accuracy and robustness of the algorithm.Fourthly, a face recognition simulation system was set up used to test the performance of the algorithms, and it proves that the algorithm has relatively high recognition accuracy and robustness.