A Novel Approach on Silhouette Based Human Motion Analysis for Gait Recognition Murat Ekinci and Eyup Gedikli Computer Vision Lab. Dept. of Computer Engineering, Karadeniz Technical University, Turkey ekinci@ktu.edu.tr Abstract. This paper 1 presents a novel view independent approach on silhouette based human motion analysis for gait recognition applications. Spatio-temporal 1-D signals based on the differences between the outer of binarized silhouette of a motion object and a bounding box placed around silhouette are chosen as the basic image features called the dis- tance vectors. The distance vectors are extracted using four view direc- tions to silhouette. Gait cycle estimation and motion analysis are then performed by using normalized correlation on the distance vectors. Initial experiments for human identification are finally presented. Experimen- tal results on the different test image sequences demonstrate that the proposed algorithm has an encouraging performance with relatively ro- bust, low computational cost, and recognition rate for gait-based human identification. 1 Introduction The combination of human motion analysis and gait recognition based human identification, as biometrics, in surveillance systems has recently gained wider interest in the research studies [1][3][6]. There has also been considerable interest in the area of human motion classification [12], tracking and analysis [4] in recent years. Those are required as initial steps in gait recognition algorithms for human identification applications [1][3]. The main purpose and contributions of this paper are summarized as follows; – We attempt to develop a simple but effective representation of silhouette for gait-based human identification using silhouette analysis. Similar observa- tions have been made in [7][8], but the idea presented here implicitly more capture both structural (appearances) and transitional (dynamics) charac- teristics of gait. – Instead of width/length time signal of bounding box of moving silhouette usually used in existing gait period analysis [10][11][3], here we analyze four distance vectors extracted directly from differences between silhouette and the bounding box, and further convert them into associated four 1D signals. 1 This work is suported by KTU (Grant No: KTU-2002.112.009.1). G. Bebis et al. (Eds.): ISVC 2005, LNCS 3804, pp. 219–226, 2005. c  Springer-Verlag Berlin Heidelberg 2005