Research On Identity Recognition Technology Based On Gait Features

As the state attaches importance to public safety issues,video security monitoring in public places plays an extremely important role.Security departments can identify pedestrians through the monitoring system.Traditional biometric recognition methods include face recognition,iris recognition and so on.However,the long distance between the camera and pedestrians,the unclear image taken,the blurred face and other issues,the traditional identification methods have been unable to meet the current needs,and gait recognition has become a research hotspot in the field of computer vision because of its advantages of long-distance recognizability and low resolution requirements.Therefore,this thesis makes a thorough study on human detection and gait recognition technology based on gait features.The main research work is as follows.(1)An improved human body detection algorithm based on ViBe is proposed.In order to improve the quality of background image and eliminate ghost phenomenon to a certain extent,the background model is established by using several continuous frames of images.The adaptive radius threshold is calculated by combining the OTSU,Background subtraction and Temporal difference,which improves the accuracy of target detection and the adaptability to illumination changes in different dynamic environments.At the same time,the accuracy of human body detection is improved by using the double judgment basis combining the aspect ratio of the target and the proportion of the area of the target area.(2)A hierarchical feature fusion algorithm based on local binary pattern(LBP)and directional gradient histogram(HOG)is proposed.Using the idea of layering,LBP features are extracted in three layers,and HOG features of each layer of LBP feature image are extracted in turn,which not only depicts the contour information of human body,but also extracts local features.Using the idea of feature fusion,the LBP features and HOG features extracted from each layer are fused.Finally,three-layer features are fused sequentially to form a new fusion feature for gait recognition.Through comparative experiments,the algorithm achieves good recognition rate.(3)Two-Dimensional Principal Component Analysis and Support Vector Machine.Aiming at the problem of high dimension of gait features and too much computation,Two-Dimensional Principal Component Analysis is used to reduce the dimension of gait features.At the same time,Support Vector Machine classifier with strong generalization ability is used to classify and recognize gait features.Experiments show that the method based on two-dimensional principal component analysis and Support Vector Machine can improve the operation efficiency under the condition of high recognition rate,and achieve good recognition results.This thesis compares,validates and analyses the above two algorithms and classification recognition methods through several comparative experiments,and proves their accuracy and effectiveness.The research results can be applied to pedestrian identification in intelligent surveillance video.