Research On Cross-View Gait Recognition Algorithm Based On Human Skeleton Features

In recent years,with the rapid development of the information and digital society,many fields have adopted biometric identification technology to protect public safety and individuals.Among them,the human gait has further improved the current biometric recognition due to the characteristics of long-distance recognition,not easy to disguise,and no need for cooperation.In the field of security,human gait plays an important role.Therefore,in the research of identity recognition,recognition based on human gait is getting more and more attention,and more and more walking recognition algorithms have been proposed and improved.This paper studies the cross-view gait recognition algorithm based on the human skeleton,and studies the walk recognition algorithm from the following aspects:(1)Improved human skeleton extraction algorithm: This article applies authoritative human body pose estimation,and the commonly used openpose algorithm is applied to the gait recognition field.The advantage of this algorithm is that it eliminates the need for complex gait video or gait sequence pre-processing.In the processing link,the human skeleton features and joint information can be directly extracted through pictures or real-time.In response to the problems in the experiment,the openpose algorithm was improved to reduce the parameters in the network,and the comparison was verified on the LSP data set with key point standards.At the same time,the skeleton of the gait sequence in the gait database is extracted,and the experimental comparison is made to fix the previous problems.(2)Conversion of gait perspective: Because the gait video or gait sequence images taken at 90 ° viewing angle are the most obvious and easy to identify.In this paper,the VTM algorithm that can convert feature vectors irrelevant to gait angles to a specified angle is used to convert gait sequence angles into 90 ° angle images,and experiments are carried out for experimental comparison.The results show that the Euclidean distance and 90 ° at each viewing angle are improved after conversion.Then use the improved openpose algorithm to extract the skeleton from the gait picture,locate the key points of the skeleton,convert thecoordinates of the joint points,calculate the joint angle from the converted coordinates,and use the change of the angle with the frame to determine the step Gait cycle,a gait cycle is divided into four periods according to the angle change boundary point,and the gait characteristics are refined as body movement characteristics.(3)Classification and recognition of gait features: The BP neural network is optimized based on particle swarm optimization(PSO),and the particle swarm optimization BP neural network is compared with the BP neural network optimized for small batch stochastic gradient descent.Excellent classifier.In order to improve the accuracy,this paper will extract the extracted gait features based on the MIV algorithm,and select the features through experiments to send them to the network.Through the comparison of experiments under different perspectives,it is found that the recognition rate of the BP neural network based on particle swarm optimization is smaller than that of the batch stochastic gradient descent optimization.Therefore,the particle swarm optimized BP neural network is selected as the classifier.Finally,the selected classifier is used for gait recognition under different cross-view covariates to verify the feasibility of implementing cross-view classification and recognition based on human skeleton,and compare it with the existing cross-view algorithms.