Research On Gait Recognition Based On The Rate Of Regional Changes And The Moment

The inherent biological characteristics of people can be utilized to authenticate their identities. Many biometric recognition technologies such as face, fingerprint, voice and iris have found successful application and offered powerful means to recognize identities exactly and guarantee information security. However, with the development of society, these identification approaches can not satisfy the requirements due to their own limitations. It is imperative to develop and research new identification methods with more effectiveness. The gait recognition is just one of these new technologies being developed for identification. It aims to authenticate the identity of people according to their posture of walking. It is one of the most potential biometric recognition technologies from long-distance. So gait recognition has gained extensive attention.For the theme of gait recognition, this thesis discusses the following issues:(1)The image preprocessing technology which is for the gait video images. An average algorithm is used to construct the new background in this thesis. Then the area of human in the image is gained by utilizing the background subtraction algorithm. Through combining the global threshold method on the rate of the regional changes and the invariant moment is proposed. At first, through calculating the rate of local regional changes in each sequence of image cycle, the dynamic characteristics of the moving target are gained. Then the static characteristics of the moving targeted and the morphological processing, the segment of motion area is achieved.(2) The extraction of the gait information on static and dynamic characteristics. An extraction of characteristics algorithm which is bas are extracted through combining the regional division of the single-frame and the invariant moment features. The effective extraction of dynamic and static information is implemented after integrating the two features.(3) In the part of classification, the support vector machine classifier based on statistical learning theory is applied. On the CASIA gait database, a series of experiments are done. Those experiments are based on different features including the rate of regional changes, the moment invariant and the fusion of above two. The results of the experiments are analyzed. The performance evaluation uses cumulative match score (CMS) and receiver operating characteristic (ROC), which respectively evaluate identification performance and validation performance.Experimental results show that the utilization of the integration of static and dynamic gait information, and the fusion of various features of gait achieve a great recognition performance. Those results also prove that the method which is proposed in this thesis is effective and can be widely applied in the gait recognition domain in the future.