Gender Classification Based on Fusion of Weighted Multi-View Gait Component Distance

In this paper, a novel fusion method for gender classification from gait based on multi-view video sequences is proposed. At the feature level, each human silhouette in a whole gait period is segmented into eight different components. Then at the match score level, the discrimination distance of each corresponding component under every camera-view angle is respectively weighted. The two-dimension weighting coefficient matrix is calculated by our presented statistical algorithm according to the expectation and variance of within- and between-class distances. A weighted sum rule is employed as the fusion scheme to finally generate the multi-view-fused discrimination distances. Experimental results show an improvement on the correct classification rate and prove our work practically meaningful for gait recognition especially in a multi-camera surveillance system.