Design And Implementation Of Face Recognition System Based On Multiple Cameras

Video surveillance,as a universal security tool,has played a significant role in combating crime and maintaining social stability.With the progress of scientific and technological civilization,intelligent video surveillance has been paid more and more attention by people.Video surveillance combined with face recognition technology can automatically identify the target people in the scene and complete the handling of specific events,which can significantly improve the intelligence of video surveillance and lead a wide application prospect.Aiming at the problem of low efficiency of detection and recognition with simple camera in a large space,a video surveillance's face recognition system based on multiple cameras is designed and implemented.The system includes face detection,face tracking,posture optimization,face recognition and other functions.High recognition rate has been acquired through test in the laboratory scenes,and the face detection efficiency with multiple cameras has also been improved.Thus high application value may be expected.The major works in this paper are as follows:1.Architecture and overall scheme of face recognition system based on multiple cameras has been proposed.A distributed face recognition system consisting of multiple monitoring computers with cameras and a background computer for face recognition calculations is built.The faces in the video acquired from the camera are detected and tracked by each monitor computer,from which a relatively better posture face is sent to the background computer by the network,where face recognition is performed centralized processing.2.Face detection function has been achieved.Adaboost face detection algorithm containing pre-selected by skin color is used to detect multiple faces.Firstly,the threshold model under HSV color space is used to segment the skin color.Then,the Adaboost algorithm is used to detect the face in the selected region of interest.The method has effectively solved the problem of high error detection rate in complex background.3.Face tracking function has been implemented.Improved CamShift algorithm is used to track multiple faces in the monitor area,and a method of image edge background subtraction is used to monitore the people passing in and out in real time.4.Face Optimization function has been achieved.Because of the posture face monitored may chang at any time,to provide a relatively better posture face image for everyone to reduce the influence of the face posture on the follow-up face recognition,a simple and not strict posture estimation method by judging the rotation degree of face is used to optimize the face.5.Improved convolution neural network is applied to face recognition.In order to solve the problem of low face recognition rate due to the influence of light,expression and other uncertain factors in the actual scenes,a simple and high efficient convolution neural network is proposed,which uses the advantages of extracting the robustness feature automatically.The convolution kernels in the first convolution layer of the network are replaced by a series of two-dimensional Gabor filters which simulate the visual cortex characteristics,meanwhile,a random partial connection method is used in some part of pool layers and convolution layers.Basing on the experiments on ORL,Yale,AR and homemade face database,the improved convolution neural network achieves the ideal effect in face recognition task,which is superior to the other classical face recognition algorithms.