Human Gait Recognition Based On One-dimensional Motion Curve On The Side View Angle

With the rapid development of technology and economy,the information security has received great attention.The status of identity authentication is becoming more and more important,meanwhile the safety factor of traditional authentication methods is getting lower and lower.The characteristics of human,that is,the individual features which exit in the human itself,are particularly important.There is great application prospect in the field of gait features of human because of its properties of long distance observation,non-contact recognition and hard camouflage.In this thesis,the human gait recognition on the side view angle is considered by analyzing one-dimensional motion curve,and human detection,feature extraction(including gait curve)and classification method are studied considering the low color contrast between foreground and background.The research work in this thesis includes:1.Human detection.The characteristics of the selected database images are analyzed,and the background is modeled by median method.Because of the low contrast between human and background color in the image,a large number of foreground pixels are left on the updated background,which makes the foreground images obtained by background difference method have a lot of holes.Therefore,an improved background model updating method based on the mix-color space is proposed,making the updated background closer to the real background.Furthermore,in order to reduce the influence of illumination mutation on human detection,an adaptive brightness adjustment algorithm is proposed,which can effectively deal with the noise interference caused by illumination changes.In addition,the OTSU algorithm is used to obtain the threshold value,when binarizing the human image,and the efficiency of the algorithm is improved.2.Gait curve extraction.Due to the periodic characteristic of gait represented in cyclical changes in human walking,the periodicity of the human gait is analyzed,and the normalization processing to the image is carried on.In order to analyze the gait features of human,the binarized human region is skeletonized and the key nodes of the human are marked out.By comparing and analyzing the different skeletonization algorithms,an improved ZS(Zhang and Suen)thinning algorithm is proposed,which makes the skeleton information closer to the central axis of the human.The trajectories of the key nodes are extracted,and according to the trajectory characteristics,models closer to the gait trajectories are obtained through different fitting methods.3.Feature extraction and classification.The feature of the curve is analyzed and the eigenvectors are constructed.By using the k-Nearest Neighbor(kNN)classifier,the similarity of different feature vectors of different people and the same person is compared respctively,which proves that the eigenvector can be effectively used in gait recognition.The k value is selected by comparing the correct recognition rate of k with different values.In this thesis,gait recognition is carried out by using the sequence of CASIA-Dataset-A from the side view angle.The experimental results show that the proposed algorithm can achieve good results.