| Human gait recognition is an emerging artificial intelligence technology, which has been getting more and more researcher’s attention. The difference of each person’s physical structure determines the uniqueness of each person’s gait and the feasibility of gait study. Gait recognition is based on different walking posture to judge the people identity. Gait is the only biological characteristics which can be remotely sensed. Generally, gait recognition technologies include moving object segmentation and preprocessing, feature extraction and gait recognition. In this paper, the main research work is as follows:(1) According to discussing for the development of gait recognition in domestic and foreign, it analysis of human moving target detection method and background modeling method. On this basis, it selects the background subtraction method to contour segmentation and preprocessing.(2) For the amounts data of gait video stream, the later algorithm complexity, lower recognition rate and other issues, it uses the key frame algorithm of gait cycle sequence to reduce system running speed. It selects the human profile aspect ratio to detecting gait cycle, and finally through the extreme value method to determine key frame.(3) Through the study of existing research results, it grasps the related description method of gait characteristics. It uses normalized Fourier descriptor method and normalized Zernike moments method to describe people contour, and using dynamic time warping algorithm to recognition.(4) Discussing the existing motion recognition algorithm, we compare the advantages and disadvantages of various algorithms. On the basis, the paper present one new gait recognition algorithm based on weighted distance, which uses to experiment on CASIA database, self-built database and MIT database.Through the large number of experimental results and data analysis showed that the novel gait recognition algorithm can accurately identify the person’s identity and their corresponding gait. |
PDF Full Text Request