Gait-based Recognition Of Humans Using Continuous Hide Markov Model

As a result of the increasing demand for security in modern society, biometrics technology is widely used in the security, identification and other areas of certification because of its special security, stability and convenience. Different from the first-generation biological features, such as face, fingerprint, iris, which are restricted to close distance detection. As a new biometrics, gait can be detected and measured under low-resolution at a distance. Also gait is hard to disguised and conceal,and it is non-invasive, therefore, gait analysis plays an important role in the visual monitoring, control, identification, etc.Gait recognition research is currently in its beginning. The term gait recognition is typically used to signify the identification of people in image sequences by the way they walk. The goal of this thesis is to investigate the information contained in the video sequences of human gait, and to perform personal identification based on the information. Focusing on this topic, this dissertation mainly includes the following issues:Firstly, preprocess the gait sequences. According to the simple background in the gait sequences we used, we analysis and compare the usual motion detection algorithms, and then use background subtraction method to detect motion; analyze the periodicity of lower limbs movement and use silhouette width information to compute the gait cycle. In the feature extraction stage, based on an intuitive thought the sufficient individual identity information can be reflected by the joint angle trajectory of body parts during the walking, this thesis puts forward a new gait representation method without modeling that we locate joint points and extract joint angle information through anglicizing the lower limb movement, at the same time we extract the width information of upper body as the accessorial feature. Furthermore we use Kalman filter to track the joint points in the occlusion sequences, The occlusion problem can be solved by the position prediction in the process of the tracking of joint points.Lastly, we do researches in the hidden markov model and analyse the feasibility of HMM, then realize an algorithm of gait-based human identification on the base of HMM. We put forward the method of getting the exemplar through analysing the joint angles of lower limbs during the different walk states, then this thesis adopt the modelling scheme that the exemplars can be considered as analogues of states of the HMM and the distance between the exemplars and the image features is used as the observation vector. The experimental results show that the proposed approach is quit robust and has a high recognition rate in CASIA database.