International Journal of Soft Computing and Software Engineering (JSCSE) Vol.4,No.1, 2014 Published online: Jan 25, 2014 e-ISSN: 2251-7545 DOI: 10.7321/jscse.v4.n1.1 1 Multi-view Gait Based Biometric System Hu Ng 1* , Wooi-Haw Tan 2 , Junaidi Abdullah 3 1,3 Faculty of Computing and Informatics,Multimedia University,Cyberjaya, Malaysia 2 Faculty of Enginnering,Multimedia University,Cyberjaya, Malaysia Email: 1 nghu@mmu.edu.my, 2 twhaw@mmu.edu.my, 3 junaidi@mmu.edu.my Abstract. This paper presents a multi-view gait based biometric system that able to work well in various walking trajectories and covariate factors such as clothing, load carrying and speed of walking. Our approach first applies perspective correction to alter the silhouettes from oblique view to side-view plane. Next, joint locations of hip, knees and ankles are estimated based on a priori information of human body proportion. Dynamic and static gait features are then extracted by the proposed extraction technique. Gaussian filter is applied to smooth the features in order to reduce the influence of outliers. Feature normalization and selection are subsequently applied before the classification process. The experiments were carried out on SOTON Oblique Database and SOTON Covariate Database from University of Southampton. From the experimental results, the proposed system achieved 92.5% and 96.0% correct classification rates for both databases respectively. Keyword: gait recognition, biometrics, covariate factors * Corresponding Author: Ng Hu, Faculty of Computing and Informatics, Multimedia University, Malaysia, Email: nghu@mmu.edu.my 1. Introduction Human gait is a unique locomotive pattern which comprises synchronized movements of body parts, joints and the interaction among them [1]. Therefore, it can be counted as a unique biometric identifier. In 1973, psychological research finding from Johannson [2] has demonstrated that human can identity walking subjects based on the light markers that are attached to their legs. From there onwards, it has inspired many researchers to work on gait analysis for human identification purpose. Gait is an unobtrusive biometric, which can be captured from a distance and without require any intervention from the user. However, its performance can be affected by covariate factors, for example illumination of light, time, load carrying, speed, clothing and camera view-point. For this reason, it makes gait recognition system as a challenging issue. Recently, many research works concentrated on the multi- view gait analysis. As in realistic surveillance scenarios, subjects are supposed to walk in various directions during the journey. In this paper, we presented a new multi-view model based gait biometric system with joint detection approach. The gait based biometric system able to compute joint angular trajectories precisely even in occluded silhouettes. It consists of four phases: (a) view normalization to handle the changes of camera capturing angle; (b) gait feature extraction to extract the required gait feature from subject silhouettes; (c) feature normalization and selection to find those positive significant gait features; (d) classification is carry out to demonstrate the performance of the propose system.