Research On Human Identification Algorithm Based On Gait Features

With the development of science and technology,great changes have taken place in people's production and life.Among them,biometric identification technology refers to the technology that uses individual biological differences of human beings to distinguish human identity with the help of scientific methods,which is also called biometrics.Existing biometrics include fingerprint recognition,iris recognition,face recognition,voice print recognition and so on.Existing mature biometric technologies based on computer vision such as fingerprint recognition,face recognition and iris recognition are strongly dependent on high-quality image information,which requires the cooperation of the identified person to collect local image information of the human body.However,in some special applications of human identification,such as judicial identification,criminal suspects will deliberately hide fingerprints,faces and other image information in the crime process,so that human identification cannot be carried out.At this time,the recognition of human gait,which has advantage of not requiring the cooperation of the identified person,difficult to hide and disguise,could be recognized by gait images,has become a hot topic in the field of identity recognition in recent years.In this paper,a identification algorithm based on gait features is studied,which aims to extract human gait features from gait images without details,and then use the gait feature to identify people.The algorithm utilizes the learning ability of deep neural network,trains and verifies the algorithm by means of supervised learning,and proves the effectiveness of the algorithm in theory and practice.Firstly,this paper introduces the research background,research significance and research status at home and abroad.Subsequently,the paper introduces the Horizontal Dropout algorithm in detail,including Sporadic Horizontal Dropout(SHD)and Consecutive Horizontal Dropout(CHD),and simultaneously analyzes the experimental results of SHD and CHD,and concludes that the CHD has better performance in extracting human gait features.Then,the design and implementation of the identity recognition algorithm is fully explained in this paper,including selected dataset,data pre-processing,the process of data-cleanup,and complete feature extraction network.The feature extraction network is divided into five parts: pre-convolutional neural network,combination strategy,Multi-dimensional Consecutive Horizontal Dropout(MDCHD)algorithm,full connection layers and distance calculation.It is proposed in this paper that the application of residual blocks in the pre-convolutional neural network to improve the network learning ability.It is also one of the innovative key technologies proposed in this paper to combine the spatial characteristics of the spatial features in the time dimension to obtain the spatial and temporal fusion feature.And one of the highlight technologies proposed in this paper named MDCHD has completed the feature level data amplification so it improves the generalization ability of the neural network simply and effectively.Meanwhile,this paper collates all the related experiments of research on human identification algorithm based on gait features.Thegordian techniques proposed in this paper are tested experimentally one by one firstly.The verification techniques including data preprocessing,pre-convolutional neural network,combination strategy and MDCHD algorithm.The experimental results show that all the proposed algorithm is effective.Finally,the experimental verification of the integratedresearch on human identification algorithm based on gait features is carried out.The three applications have been applied which are cross-type gait recognition,cross-view recognition and gait-based re-identification.The experimental results of first two applications have been compared with the state-of-the-art algorithms with same training set and test set.The result of comparison illustrate that our method has more advantages of gait feature extraction and generalization ability.See from the experimental results,our method has outperformed the state-of-the-art approaches in almost all walking types and views.Besides,we developed inference platform and designed inference test with human gait video out of CASIA-B dataset,which proved that our method also performs grate in practice.