Research On Identity Recognition System Based On Face And Gesture Dynamic Password Recognition

In recent years,identity recognition based on face recognition technology has been widely used,but there are still many shortcomings in the identity recognition system based on face only.For example,if it is attacked by means such as photos,videos,masks,etc.,there are hidden security risks.Therefore,to cooperate with other methods for identification,dynamic password is one of the very effective methods.However,in some special moments or occasions such as the outbreak of the new crown virus,the traditional touch screen,keyboard and other input methods are very likely to cause virus transmission.Therefore,this paper proposes the use of digital gesture recognition technology to complete the dynamic password input.The following specific instructions:This article first analyzes the market demand of the system,determines its application occasions,and proposes the overall design scheme of the system.Its process is: first,face recognition,and then pass the dynamic password to the mobile phone of the person to be verified.The person to be verified makes corresponding gestures according to the obtained password to complete the authentication.On this basis,the implementation algorithms of face detection,face recognition and gesture recognition are studied respectively.For the face detection problem,the MTCNN deep convolutional neural network structure is used to implement,but the original MTCNN algorithm has poor real-time performance on the CPU.In this paper,the deep separable convolution technology is used to replace the traditional deep convolution algorithm,which reduces the parameter calculation and improves the realtime performance.In face recognition,based on the Mobile Face Net network,the CBAM module is introduced into the original bottleneck layer,which enhances the network's feature extraction ability.And for solving the problem of the original loss function hyperparameters needing manual debugging and testing,the introduction of Adacos loss function realizes the adaptive dynamic adjustment of hyperparameters and improves the recognition accuracy.In terms of gesture recognition,this paper improves the YOLOV3 algorithm to adapt it to this research system,uses the K-means clustering method to generate a new preset frame,and replaces the traditional feature extraction network with depth separable convolution.Meanwhile,we introduce the CBAM module in order to improve the network's feature extraction capability and the accuracy of detection.In addition,this paper conducted a comparative experiment to verify the effectiveness of the improved method.Finally,according to the proposed algorithm and the overall system plan,the system functions are specifically designed and implemented.In addition to real-time identification,the functions of the personnel information database are also realized.