Research On Gait Recognition Algorithm Based On Deep Learning

Gait recognition algorithm based on convolutional neural network(CNN)can verify the identity of the target by extracting the appearance contour features of the human body when walking.Because of the advantages of long recognition distance,no contact and high recognition efficiency,it has attracted more and more attention from researchers in recent years.Compared with traditional biometric technology,it has a broader application prospect in public safety,medical and health fields.In real scenes,the gait information of the target is affected by different factors,such as the change of ambient light,the change of wearing and viewing angle,and the difficulty of deploying CNN model,which reduces the accuracy of gait recognition and limits the popularity of related algorithms.This paper proposes a gait recognition algorithm based on lightweight CNN.The main research contents are as follows:(1)Introducing channel attention mechanism to improve the efficiency of extracting key feature information from convolution layer,and generating redundant feature maps through linear operation to reduce the parameters in convolution process.In order to avoid the distortion of feature map and the over-fitting of deep network model in the training process,a lightweight module is designed to imitate the structure of residual network,and a new L-ResNet-50 CNN model is built by stacking.Experimental results show that compared with the traditional model,the new network model can effectively improve the gait recognition rate when the target carries a satchel or wears a coat,and the computation and training time of the model are greatly reduced;(2)In order to solve the problem of gait recognition with small sample set and cross-viewing angle,a dual-channel feature matching network model is designed in this paper.The global features of the gait energy map of probe and reference set are extracted by two channels,and then the SVM loss function is used for binary classification matching.Experiments show that the model can obviously improve the accuracy of cross-view gait recognition under complex conditions;(3)Because of the lightweight CNN model,this paper designs a gait recognition system based on deep learning,which can be deployed on embedded development board.The system uses the deep learning framework on Tensorflow to build a CNN model,which is deployed on NVIDIA Jetson Nano deep learning card board.The system includes three modules: Target acquisition,image processing,and gait recognition and analysis,which can complete gait recognition in simple environment.