The Research On Gait Recognition Based Gait Energy Image And Weighted Mass-vector

Gait recognition aims at discriminating individuals by the way they walk. Compared with other biometrics such as face, iris, fingerprints, etc, gait recognition has its own advantages such as data acquisition from distance, non-invasion. Hence, it has great potential for video surveillance and human behavior analysis.Gait recognition is a blend of multi-disciplinary research fields, which includes segmentation, feature extraction, feature processing, pattern classification and gait database content. In order to improve the recognition rate and meet the real-time requirements of gait recognition, we explore and research on gait recognition from image processing and pattern recognition. The main work and achievements are as follows:In order to improve the accuracy of gait feature data, we propose a silhouette denoising method based Gait Energy Image(GEI), which repairs single-frame image missing by the main part of silhouette obtained by doing threshold filter to GEI.In order to solve the problem of high computational complexity and maybe no solution of current feature processing method, we propose Weight Vector to describe the contribution of each element of gait feature. In addition, with the increasing number of samples, Weight Vector can dynamically adjust to extract the relatively stable features more accurately, which is expected to get the best recognition rate to future test samples.we propose a gait recognition method based on weighted mass vector, which adds the denoising processing to the initial silhouette and extracts the mass-vector as gait feature, then introduces Weight Vector to the traditional Normalized Euclidean Distance(NED) for similarity measure, finally uses the nearest neighbor method for classification.We do experiments on the CASIA database provided by Institute of Automation, Chinese Academy of Sciences, compare feature processing based Weight Vector and traditional method, also we compare the experiment results with and without silhouette denoising processing based GEI. The results show that the silhouette denoising method improves the accuracy of gait feature data and the feature processing method based on Weight Vector has high real-time performance and classification ability. Recognition rate in the 0°and 45°degrees are 98.75% and 97.5% respectively.