Research On Gait Behavior Analysis And Alignment

At present,the social environment is increasingly complicated,and all kinds of cases are high incidence,which affected the social harmony and stability.In order to deal with the security environment,the intelligent video surveillance is gradually improved and play a monitoring role.In the field of public security,people are the main goal and gait feature is one of the important biological information,which can be used to target suspects.Thus,gait recognition technology will play an important role in investigating suspects and increasing warning time and automatic identification.The gait behavior analysis and alignment method are researched in this thesis,including key technologies such as moving human target detection,gait feature extracting,gait feature fusion and classifier design and so on;designing and implementing the gait analysis software,which can detect moving target in monitor,extract gait feature and classifier recognition.In terms of the detection of the moving human target,the moving human target detection algorithm based on Gaussian Mixture model are researched and implemented.This model updates background by adjusting parameters,and get clear target silhouette.Simulation experiment with video having light changing and video came from gait database,and the result indicated that the model is better than background subtraction.In terms of the extraction of gait feature,improvement is done on gait period based on swing distance of lower limb.At first,locate the position of centroid,then calculate distance from centroid to pixels of below the knee joint,and this way can fall the effecting of swing arm;extract of features of joint angle based on pendulum model,and extract gait energy image based on statistic,and use the principal component analysis to make dimension reduction.In terms of the feature fusion and the design of classifier,and this thesis improve a method of vector construction of Multi features fusion.It constructs vector with period,joint angle and GEI by weighting.Gait recognition is based on minimum Euclidean distance classifier.Simulating experiment with CASIA database,the result indicate that recognition rate of feature fusion is 84.37%,which is higher than others.In terms of the design and implementation of the gait analysis software,this thesis designs the software which is based the platform of Matlab,using Matlab GUI designing the interface.And it includes 3 modules of the detection of the moving human target,the extraction of gait feature and classifier training and recognition.