Design Of Entrance Guard And Attendance System Based On Face Recognition

As a biometric recognition technology,face recognition has the characteristics of universality,high reliability and strong individual differences.It has broad application prospects in the field of smart security.According to the needs of the access and attendance control system in the school campus construction,this thesis implements the face recognition algorithm based on deep learning to the face recognition access and attendance control system,and lightly improves the algorithm for the problem of large computational complexity.The face recognition task is divided into three processes: detection,alignment,feature extraction and classification.This thesis proposes a lightweight improvement scheme for detection and feature extraction algorithms,and implements DAN(Deep Alignment Network)key point detection and affine transformation to achieve alignment.From the perspective of the actual application of access and attendance control system,this thesis implements the design of live detection,face tracking and access attendance control modules.The main work includes: 1.In face detection,replace the VGG-16 backbone network in the SSD(Single Shot Multibox Detector)detection framework with a lightweight MobileNet network,MobileNet network structure is simple and use a deep separable convolution structure to implement convolution operations.In the case of ensuring the accuracy,the amount of calculation can be greatly reduced,thereby reducing the computational complexity of the detection model.2.In the face alignment,the DAN algorithm with good robustness to the face pose is used to realize the key point detection,and the affine transformation is used to realize the face alignment.3.In the aspect of face recognition,a lightweight feature extraction scheme is proposed.MobileNet is used as the backbone network to reduce network parameters,reduce the complexity of model training and improve the running speed,this thesis proposes the Center_Softmax joint loss function,which increase the inter-class spacing and reduce the intra-class spacing,and the network output is used as the feature vector of the input sample.4.In the implementation process of the face recognition access and attendance control system,the method of detecting blinks is used to prevent the photo fraud problem;the method of "single frame detection,multi-frame tracking" is used to solve the problem of video jamming in real-time face detection and recognition;using the Raspberry Pi + "Server Chan" + WeChat to accomplish the access and attendance control system.According to the test results of the detection and recognition algorithm in the public data set and the real scene,the improved detection algorithm has a great increase in speed compared with the original SSD framework on the premise of less loss of accuracy,and Compared with the FaceNet framework,the speed of the feature extraction algorithm is improved by more than 30%.The recognition system has more than 95% recognition accuracy in various scenarios,and it is robust to changes in lighting,angle,background and other factors,and has achieved the established functional and performance indicators.