Cross-view Gait Recognition Based On Image Sequence

Gait recognition technology is an important research direction in the field of computer vision.It is a method of identifying the identity of the target by analyzing the walking posture of pedestrians.Compared with the traditional biometric identification technology,it is difficult to camouflage,requires no cooperation,and can be observed from a long distance.Gait recognition technology is gradually playing an important role in public security,smart city construction and other fields.However,in the actual scene,the occlusion of the pedestrian’s body or the change of the observation angle will lead to deviations in the extracted gait features,then affecting the recognition results.At the same time,due to the large number of network layers and the large amount of computation in Gait Set,the gait recognition system is difficult to deploy in lightweight monitoring equipment.In order to solve the above problems,this paper focuses on the cross-view gait recognition which is based on image sequences.The specific work is as follows:(1)A gait recognition method based on original sequence-level features and joint loss function is proposed to improve the recognition accuracy of the model.First,the gait energy map is added to the MGP module of the Gait Set network as the original sequence-level feature,which makes up for the problem of the lack of global information in the deep network;secondly,the Softmax loss function is added to jointly supervise the training of the network.During network training,it can increase the distance between sample categories,reduce the distance within sample categories,and alleviate the problems of slow convergence and unstable performance of the Triplet loss function.the experimental results show that under the conditions of NM,BG and CL walking,the correct rate of gait recognition have been improved0.2%,3% and 4.7% respectively.(2)A gait recognition method based on depthwise separable convolution and SE modules is proposed to construct a lightweight network.First,the conventional convolution in the network is replaced with a depthwise separable convolution to realize the compression of model parameters;secondly,the SE module is applied to the backbone network,so that the network can learn the importance of each channel to the task independently;the experimental results show that the training time of the network is shortened by 52.5%,and the storage space occupied by the model file is reduced by 19%.(3)A gait recognition method based on depthwise separable convolution and SE modules is proposed to construct a lightweight network.First,the conventional convolution in the network is replaced with a depthwise separable convolution to realize the compression of model parameters;secondly,the SE module is applied to the backbone network,so that the network can learn the importance of each channel to the task independently,and improve the network’s expressive ability to understand information;the experimental results show that the training time of the network is shortened by 52.5%,and the storage space occupied by the model file is reduced by 19%.