Research On Cross-view Gait Recognition Based On Spatial-temporal Features

In recent years,with the rapid development of science and technology and information society,the impact of information security on people is also growing.Many application scenarios gradually use biometric identification technology to maintain public and personal security.Compared with other biological features,human gait has the advantages of long-distance recognition,difficult to camouflage,no need to cooperate and so on.However,the research of gait recognition is greatly affected by the factors of visual angle,clothing and belongings.Therefore,the research work of this paper is to improve the recognition performance of gait recognition algorithm under the above conditions(1)Aiming at the problem that the accuracy of gait recognition will be reduced when the clothing and carrying items are changed,this paper proposes a novel method to treat human gait as a set of unordered sets and a set of ordered sequences at the same time.The innovation of this method lies in that it is different from most other methods that only extract gait space or temporal features.By using the difference of human appearance and the difference of human joints with time,the fusion of temporal and spatial features can reduce the impact of changes in clothing and carrying items.(2)Based on the above ideas,a network structure called GSTSN(Gait Set temporal spatial network)is constructed based on the temporal and spatial characteristics of gait.In the proposed network structure,the deep convolution neural network is used as the spatial feature extractor,and the long short-term memory network is used as the temporal feature extractor.Before extracting temporal features at the same time,an image dimensionality reduction operation is constructed to achieve dimensionality reduction while preserving image temporal features.The horizontal pyramid structure is introduced to fuse the extracted gait temporal and spatial features to form temporal and spatial features,which are used as gait features for classification and recognition to improve the accuracy of gait recognition.(3)In order to verify the effectiveness of temporal and spatial features,comparative experiments are carried out on CASIA-B data set.The designed experiments include gait contour image experiments and human gait image experiments.Other gait recognition algorithms are based on one of the image types.The experiments designed in this paper can better evaluate the performance of gait recognition algorithms.The experimental content includes not only basic experimental comparison,but also complex experimental comparison such as limited frame number,multi view and multi gait conditions,so as to illustrate the performance of gait recognition algorithm in practical application.Experimental results show that the proposed method can extract more abundant features of human gait with a small amount of computation,which not only improves the recognition rate,but also has a certain robustness to the changes of clothing and carrying items.At the same time,based on CASIA-B data set,a more complex practical comparative experiment is designed.The experimental results show that the features of gait temporal information and spatial information are more suitable for real gait recognition tasks.