摘要：Human gait recognition aims to identify people by their walking style. In this paper a difference image based human gait identification method is proposed. For each human gait images sequence, gauss model based background estimation is used to segment frames of the sequence to obtain the silhouette images with less noise. By comparing the difference of two adjacent silhouettes in the images sequence, we can get a difference images sequence. Every difference image in the difference sequence indicates the body moving feature during ones walking. By projecting every difference image to Y axis or X axis we can get two feature vectors. Project every difference images of the whole walking images in one walking cycle we can get two matrixes. These two matrixes indicate the style of ones walking. Then Two-Dimensional Principal Component Analysis(2DPCA) is used to transform the above matrixes to two vectors while at the same time keep the maximum separability. Finally the similarity of two human gait images is calculated by the Euclidean distance of the two vectors. Experimental results on a gait database of 124 people show that the rank 5 identification accuracy can achieve 92%.
