Gait recognition based on dynamic region analysis

Gait Energy Image (GEI) has been proved to be an effective identity signature in gait recognition. But previous approaches only treat this 2D image representation as a holistic feature and neglect the intrinsic dynamic characteristics of gait patterns. In this paper, we use variation analysis to obtain the dynamic region in GEI which reflects the walking manner of an individual. Based on this analysis, a dynamics weight mask is constructed to enhance the dynamic region and suppress the noises on the unimportant regions. The obtained gait representation called enhanced GEI (EGEI) is then represented in low dimensional subspace by Gabor-based discriminative common vectors analysis. We test the proposed approach on the USF HumanID Gait Database. Experimental results prove its effectiveness in terms of recognition rate.