Study Of Method For Gait Recognition Based On The Fusion Of Angles And Body Silhouette Feature

With a growing need for security in modern society, biometrics recognition, as a human identification approach for access control in security-sensitive environments, has been greatly researched and developed. To operate successfully, the established biometrics such as face,fingerprint usually require proximal sensing or physical contact. However, Gait is the only one biometric that can be easily perceived at a distance. So, gait recognition, which aims to identify individuals by their walking manners, is very attractive in the field of visual surveillance.At first, gait sequences are preprocessed. By analyzing and comparing kinds of motion detection methods, and considering the simple background of gait sequences, background subtraction is used in gait detection. Gait cycle is analyzed, then width and height of body analysis is performed to computer it.Next, based on the idea that joint-angle trajectories of body parts in walking motion include sufficient dynamic identity information, a gait recognition method based on lower-limb motion analysis and dynamic time wrapping is proposed. For each gait sequence, according to the knowledge of human body anatomy, the coordinates of lower-limb joints are obtained by analyzing lower-limb motion, and then the trajectories of lower-limb angles in one cycle are extracted as feature vectors. Dynamic time wrapping is used to measure the similarity of different sequences, then, nearest neighbor algorithm and K-neighbor algorithm are finally performed to realize gait recognition.At last, in order to solve the problem that most gait recognition methods based on single feature can not get satisfied recognition results, according to the idea of feature fusion, a gait recognition method using fusion of lower-limb angles and body silhouette at score level is proposed. Each feature is assigned to weights, which can make them combine in suitable proportion. Experimental results demonstrate that the recognition rate of proposed method is 95%, much higher than those based on single feature.