With the development of the security requirement of modern society, it becomes the mainstream to identify one's identity with biometric today. Among the different biometrical characters, gait has its own advantages which is untouched and requires little about the quality of the original data, such as image resolution. The former advantage makes the gait's input to be non body-invading, and the latter broaden its applications. Furthermore, gait is difficult to be disguised. All of these lead the gait recognition becomes the second-generation biometric recognition technology based on the vision movement. Also, many researches have been deployed on this field.Working on gait recognition methods, this dissertation concentrates on the following topics:The status-art of gait and some background knowledge of principal curves are introduced. Considering kinds of detection methods, the background subtraction is used in detection. On the other hand, principal curves can reflect the inherent structure of the data and describe nonlinear data, which is beneficial to the contour extraction. So the dissertation proposes a silhouette contour description based on K principal curves. Experiments prove that the K principal curves can sketch silhouette contour accurately.Analyzing the description ability of silhouette contour, the dissertation proposes a novel gait representation based on contour matrix. The representation utilizes Kronecher product to get the contours' position differences, which are gait features. It contains both the detail static shape of silhouette and their dynamic information, therefore represents effectively the spatio-temporal pattern of gait.Because the linear principal component analysis is not adapted to nonlinear data, a new nonlinear analytic method (principal curve component analysis) is presented. This method analyzes the data in nature, emphasizes the non-parametric characteristic and models nonlinear data effectively. In addition, the dissertation considers principal curves as the new method of classification, and then defines the new comparability measurement and classification rule. Different