The Design And Implementation Of Real-time Gait Recognition System Based On Deterministic Learning

Gait recognition is a new biometrics recognition technology, it aims to recognize peopleor detect physiological, pathological and mental characteristics by their walking style.Compared with the other biometric technology, gait recognition has a potential to identify bythe remote video or low-quality videos, and gait is hard to hide or disguise. With theimproving need of monitoring in the stations, banks, airports and other sensitive occasions,computer vision researchers take great interest in remote identity authentication system. Gaitfeature is the only one biometrics feature can be obtained at a distance, so from the point ofintelligent video surveillance, gait recognition will be widely used. Gait recognition hasbecome a new research area.The deterministic learning (DL) theory proposed in recent years is a new machinelearning theory, which provides systematic design approaches for knowledge acquisition,representation, and utilization in uncertain dynamical environments. Locally-accurateidentification of the gait system dynamics is achieved by using radial basis function (RBF)neural networks (NNs) through deterministic learning. The obtained knowledge of theapproximated gait system dynamic is stored in constant RBF networks. A bank of estimatorsare constructed using constant RBF networks to represent the training gait patterns. In the testphase, by comparing the set of estimators with the test gait pattern, a set of recognition errorsare generated, and the average L1norms of the errors are taken as the similarity measurebetween the dynamics of the training gait patterns and the dynamics of the test gait pattern.Therefore, the test gait pattern similar to one of the training gait patterns can be rapidlyrecognized according to the smallest error principle.This dissertation has carried on the exploration and research of gait recognition system inpractical application, based on the previous research achievements. The main contribution andinnovation of this dissertation are summarized as follows:(1) A real-time gait recognitionsystem is designed. After solving the problem of gait feature extracting automatically byOpenCV and computing rapidly by parallel computation, A real-time gait recognition systemis designed based on MATLAB GUI. We tested the system under the environment constructedby ourselves, and verified the practical application value of the system.(2) Using the Kinectdepth camera for gait recognition. We obtained the skeletal information by the Kinect depthcamera in certain scope. This dissertation tried to use the information in gait recognition, andhave been well justified based on side view direction and front-view direction. It is a attemptto solve the problem of gait feature extraction under complicated background.