Gait Recognition Using Multi-Bipolarized Contour Vector

Gait recognition has recently attracted increasing interest from the biometric community. In this paper, we propose a simple yet powerful new feature called multi-bipolarized contour vector (MBCV) for gait recognition. The proposed MBCV feature consists of four components: (1) the Vertical Positive Contour Vector, (2) the Vertical Negative Contour Vector, (3) the Horizontal Positive Contour Vector, and (4) the Horizontal Negative Contour Vector. We furthermore develop a gait recognition system based on the proposed MBCV feature. The system consists of three steps: image preprocessing including background subtraction and silhouette normalization, extraction of the MBCV feature, and classification. To reduce the dimensionality of MBCV, we use principal component analysis (PCA). To solve the classification problem, we use the Euclidean distance and a nearest neighbor (NN) approach. Finally, we fuse the proposed gait features at all levels to improve recognition performance. The proposed recognition system is applied to the well-known NLPR gait database and its effectiveness is demonstrated via comparison with previous works.