Human Gait Recognition Using Template Features With Opponent-Motion Model

 In this paper, we apply a filtering scheme to encode motional information from style of human walking into single template. This filtering process utilizes bio-inspired basis kernels which are constructed from two separable spatial and temporal basis functions. To extract opponent motions from gait in a walking cycle, wefusethese kernels together to have spatiotemporal kernels and thenconvolve itwith gait stancesThis paradigm has been repeatedineach frame of gait in a cycle and motional information is highlighted as salient regions inthe sequence of images. To make gait features (salient regions)distinctive enough within different individuals, the responses of gait filtering are aggregated together to get final Gait Salient Image (GSI). Our proposed model provides more representative gait feature since conventional gait templates such as Gait Energy Image (GEI) does not rely on any motion model. Extensive experiments on popular gait databases reveal that, our GSI-based features provides more competitive performance compared with recent published gait recognition approaches with efficiency and accuracy