Research Of Algorithm For Gait Recognition Based On Feature Fusion And Svm

Gait recognition is a biometric identification technology that identifies individuals by their walking manners. As a new physical feature, gait has many advantages, such as recognition at a distance, non-invasive, perceivable, hard to conceal, even, it is the only feature that can be captured in low resolution. So, gait recognition has been a pop research subject in computer vision and pattern recognition.The processing procedure of gait recognition consists of three parts: mobile image object extraction, feature extraction and feature classification. Our research priority is feature extraction. In this paper, two new gait feature extracting algorithm are suggested. The first algorithm is based on the fusion of spatial feature and frequency feature, the other one is based on the fusion of gait contour wavelet moment and thigh double triangle feature.Spatial feature and frequency feature are all valid in gait recognition, so, it will enhance the correct classification rate and robustness of gait system by using these two features. Based on this idea, we suggest a gait feature extracting algorithm based on the fusion of spatial feature and frequency feature. Firstly, preprocess the human gait image, including background modeling, mobile object extraction and morphological operation.; Secondly, a equidistant slicing curves based on system of polar coordinate is designed to slice the mobile object, and the slicing vector is used to describing the spatial feature; Thirdly, the slicing vector is converted into frequency signal by Fourier transform to extract the frequency feature. Finally, fusing the above two features, and then using the SVM to identify the feature.There are dynamic information and static information in gait sequence, and it will enhance the gait system's performance by fusing these two kinds of information sufficiently. So, we suggest a gait feature extracting algorithm based on the fusion of gait contour wavelet moment feature and thigh double triangle feature. Firstly, extract the key frame in a gait cycle and normalize the edge of the gait contour by analyzing the gait image after preprocessing; Secondly, due to wavelet moment is invariant to translation,scale change and rotation, we compute the wavelet moment of the contour and use it to describe the static feature of the gait image; Thirdly, to describe the dynamic features of the gait image, we build two thigh triangles, the first one consists of the hip's middle point and two knees, the other one consists of the hip's middle point and two ankles. Finally, fusing the above two features, and then using the SVM to identify the feature.The experiment results on CASIA gait database show that the above two algorithm have high correct classification rate and stability.