Pose-Based Boundary Energy Image for Gait Recognition from Silhouette Contour Information

Gait is a biometric feature that refers to walking style, and medical studies claim that any individual has an unique gait pattern. In this work, we consider a gait recognition scenario in which at least a complete cycle of gait is captured for each person and propose a new feature termed as the Pose-based Boundary Energy Image (BEI) that captures the dynamics of gait at a high resolution by deriving features from the silhouette contour information corresponding to the fractional parts of a gait cycle. The use of silhouette contour information for feature extraction is advantageous in the sense that it makes the extracted gait features more discriminative due to eliminating the redundant silhouette-level information, as used in most traditional approaches. We reduce the dimension of the Pose-based BEI feature by applying PCA, and next carry out the final phase of LDA-based classification using the reduced feature set. Evaluation of the proposed approach has been done on CASIA B and TUMGAID data sets and satisfactory results are obtained. Comparative study with existing techniques also shows that our approach outperforms the existing methods.