A new Kinect-based frontal view gait recognition method via deterministic learning

In this paper a new method for frontal view gait recognition is presented. Joint angles data are captured by Microsoft Kinect. These frontal view gait features represent the motion trajectories and gait dynamics of human walking, which are suitable for persons identification. In the training phase, gait dynamics represented by joint angle features are approximated by radial basis function (RBF) neural networks. In the recognition phase, by comparing a set of constructed estimators with a test gait pattern, a set of recognition errors with L₁ norms are generated. By using the smallest error principle, the test gait pattern similar to one of the training gait patterns can be recognized. Finally, several experiments are carried out on the self-constructed Kinect gait database, which includes at most 27 subjects, to verify the effectiveness of the proposed method.