Gait Recognition Based on Fast Shape Matching

This paper proposes a novel gait recognition method based on fast shape matching algorithm. First, we propose a simple but effective gait cycle estimating method and analyze the periodic property of gait sequences. Each frame in the sequence is assigned a phase. Those frames with the same phase are averaged to clean noises. As a result, the original long sequence with dozens of frames is simply represented by a shorter representative sequence with half of cycle length. Shape context descriptor is introduced to depict the distribution of the boundary point of each frame. In order to decrease the computational cost of point matching, we propose a fast shape matching method. The computational efficiency is at least upgraded by two orders of magnitude. The experiments on NLPR database verify the effectiveness of the proposed method. Moreover, we also discuss the influence of the number of the most similar frame and the number of sample point.