Fast Detection And Recognition Of Human Face In Video Surveillance System

A study on the technology of automatic face detection and recognition in video surveillance system is presented in this dissertation. As it has widely applied prospect in the domain of personal identification, human-computer interface, image retrieval, visual surveillance, etc, face recognition technology has been a research focus on machine vision and pattern recognition. However, because of the complicated pattern and the frangibility of human face, most face recognition methods have the weakness of large computation, low efficiency and low robustness. After analyzing of the existing methods, this paper proposes some modified approaches aim at the real time of video surveillance system and the randomicity of illuminated environment, and achieves some valuable conclusions as follows: Base on studying the methods of motion detection, a local-statistical-filter algorithm is proposed. Compared with common filter algorithm, it has the superiority in computing speed and capability of eliminating noise, and it is fitter for extracting the movement information from image sequence. After analyzing the color spaces and skin chrominance model in human skin segmentation, a method of correcting skin color in highlighted area based on adjacent area is proposed, which can resolve the problem of leaving out in skin detection because of highlight. A method of fast face feature localization is implemented with geometrical feature and gray feature. The pattern in the face recognition phase is modified. Before using the method of combining PCA and LDA to recognize face, the pattern of whole face is replace by folded face. Experimental results show that with the folded face pattern, the recognition speed is accelerated and the recognition rate is improved because of better robustness to light.