Biometrics makes use of the physiological or behavioral characteristics of people to authenticate their identities. With a growing need for a full range of visual surveillance and monitoring systems in security-sensitive environments such as banks and airports, human identification at a distance has recently gained increasing interest from computer vision researchers. To operate successfully, the established biometrics such as face, fingerprint and iris usually require proximal sensing or physical contact. However, they are hardly applicable at a distance. Fortunately, gait, the way people walk, is still visible and can be easily perceived unobtrusively. So, from a surveillance perspective, gait is a very attractive modality.

        Gait recognition is a relatively new research direction. It aims to seek distinguishable variations between the same actions of walking from different people for the purpose of automatic identity verification. Focusing on this topic, this dissertation mainly includes the following issues:

a) The establishment of NLPR gait database aims to provide data for the development of gait recognition algorithms. All image sequences are captured in outdoor environment with natural light by a surveillance camera of Panasonic NV-DX100EN Digital Video Camera. It includes 20 different subjects, and three views, namely lateral view, frontal view and oblique view with respect to the image plane. The captured 24-bit full color images of 240 sequences have a resolution of 352*240. This dataset has been shared with all related researchers.

b) Based on the idea that a specific appearance model can be learned from spatial-temporal motion pattern of gait, we propose a simple and efficient gait recognition algorithm using statistical shape analysis. For each image sequence, an improved background subtraction procedure is used to extract moving silhouettes of the walker from the background. Temporal changes of the detected silhouettes are then represented as an associated sequence of complex vector configurations in a common coordinate frame, and are further analyzed using the Procrustes shape analysis method to obtain an eigen-shape as signatures. This method does not directly analyze the dynamics of gait, but implicitly uses the action of walking to capture the structural characteristics of gait, especially the biometric shape cues. Experimental results demonstrate that the proposed algorithm has an encouraging recognition performance.

c) Based upon an intuitive consideration that recognizing people by gait depends greatly on how the silhouette of human body changes over time, we present a non-parametric gait recognition method using PCA（Principal Component Analysis）. For each image sequence, a background subtraction algorithm and a simple correspondence procedure are first used to segment and track the moving silhouettes of a walking figure. Then, eigenspace transformation based on the PCA is applied to 1D time-varying distance signals derived from a sequence of silhouette images to reduce the dimensionality of the input feature space. Supervised pattern classification techniques are finally performed in the lower-dimensional eigenspace for recognition. This method implicitly captures the structural and transitional characteristics of gait. Extensive experimental results on outdoor image sequences demonstrate that the proposed algorithm has an encouraging recognition performance with relatively low computational cost.

d) Based on the idea that joint-angle trajectories of body parts in walking motion include sufficient dynamic identity information, we propose a model-based gait recognition approach. A model-based approach together with human body model, motion model and motion constraints is first presented under a Condensation framework to track the walker in monocular sequence. From the tracking results, we can easily recover motion trajectories of main lower-limb joints such as thighs and knees. These trajectories are normalized with respect to structure and time, and are used as dynamic information of gait for recognition. Experimental results show that this approach is satisfactory.

e) Based on the fact that body biometrics includes both static appearance information of human body and dynamic motion information of walking action, we propose a personal recognition algorithm using fusion of static and dynamic cues of body biometrics at a decision level. Experimental results using different combination rules show that both identification and verification performance after fusion are improved in comparison with any single modality.