Realization Of Access Control System Based On Face Recognition

When face recognition technology is applied to the access control system,the recognition rate is easily affected by the light intensity.It is difficult to collect the image to meet the feature recognition when the light intensity is low.Suppose a single frontal image is used as the image in the face database.In that case,the phenomenon of identification error will appear in the subsequent recognition process because of the influence of the acquisition Angle.In addition,face recognition access control will also have the problem of fake face attack.Given the above issues,this paper optimized facial contour and position judgment methods,face recognition,and liveness detection and solved the above problems well.Given the influence of light intensity on face contour and position judgment,Adaboost face contour and position judgment method and LBPH feature description method with robustness to light intensity were used to combine the LBPH feature of the equivalent model and the LBPH feature of the circular equivalent mode.The detection speed was ten times faster than that of the Harr-like Adaboost method.Given the impact of single frontal image on recognition,the PCA face recognition method is adopted to expand the single frontal face image and enrich the face database.Compared with the traditional PCA face recognition method,the recognition rate of the improved method improved by10%.A convolutional neural network-based method with high accuracy is adopted to solve fake legitimate human face attacks.Embedded computing power and memory are limited,so a Convolutional neural network cannot be used efficiently.Combined with parameter sharing,dense connection,and grouping convolution method,a lightweight convolutional neural network model GDSCpeleenet with significantly reduced parameters and almost no loss of accuracy is designed.Through liveness detection experiment verification and analysis,compared with the baseline model,this model reduced the number of parameters to 30.2%,shortened the detection time by half,improved FAR by 2%,and FRR was about 18%.To solve the problem that access control cannot be opened when the server fails,online and offline modes are designed.Raspberry Pi is used as the development board,and hardware devices such as a camera and electromagnetic lock are selected to realize the face recognition access control demonstration system with the functions of face recognition,live detection,and attendance checking.The performance test of each function of the demonstration system was carried out through the face data set to verify the effectiveness of each function.The designed access control demonstration system can run stably and effectively in both online mode and offline mode during the whole test.