Research And Implementation Of Attendance System Based On Face Recognition

With the rapid development of deep learning,face recognition technology based on convolutional neural network has been widely used in all walks of life.Although convolutional neural network has high accuracy in classification,detection and other tasks,the amount of parameters carried by the network itself is huge,which leads to the consumption of large computing power for network reasoning and cannot be applied to embedded devices with small computing power,Therefore,this paper implements a face recognition system based on lightweight convolutional neural network that can be deployed in embedded devices.The research work of this paper mainly includes the following three aspects:(1)In terms of face detection,the multi-scale feature pyramid model is used to extract face features,and the deep separable convolution is used for network slimming.The MobileNetV2 network is used as the backbone network to realize face detection,which is deployed on the Jetson Nano embedded device.The experimental results show that the image resolution is 320×240 on the face detection dataset WIDER FACE,the speed of face detection is 31FPS.(2)In face recognition,the transformer mechanism based on self attention is introduced and compressed in its channel direction to realize lightweight.Combined with the advantages of MagFace loss function,the lightweight face recognition network is realized and deployed on the Jetson Nano embedded device.The experimental results show that on the face recognition data set LFW,the accuracy is 98.65%and the speed of face recognition is 34FPS.(3)An attendance system based on face recognition is designed and implemented.The face input is completed on the Android side,the face recognition is realized on the embedded device Jetson Nano,and the face recognition results are pushed to the Android device in real time.The system test results show that when the number of registrants is 500,the average time from face detection to face recognition is about 94 milliseconds,which meets the requirements of the attendance system.