Gait Recognition at a Distance Based on Energy Deviation Image

Silhouettes extracted from the videos collected with complex background at a relatively far distance are generally of low quality. Gait Energy Image (GEI) has been reported as a good feature robust to silhouette errors and image noise, but it ignores some gait motion information. This paper proposes to generate energy deviation image (EDI) based on the differences between the GEI and the silhouettes of a subject. It effectively extracts the motion information from human gait. Zoom distance is utilized to calculate the weighted combination of EDI distance and GEI distance. Nearest neighbor classifier is adopted to recognize subjects. The proposed algorithm is evaluated on USF dataset, and the performance is compared with the baseline algorithm and two other new algorithms. Experimental result shows that the proposed algorithm achieves higher overall recognition rate then the other algorithms.