Approach for tracking human being in surveillance videos

Abstract

Tracking a moving object in image sequences from a stationary video camera is a crucial task for surveillance applications. This paper proposes a hybrid technique that combines Kalman Filter (KF) and the Support Vector Machines (SVM). First, the moving target is determined according to the user's interest, and then the system state model of the KF algorithm is constructed. Second, a set of patterns are generated around the target's position. For every pattern, the Histogram of Oriented Object (HOG) is calculated to be classified into positive (humans) and negative patterns (other object) by the SVM algorithm. Finally, the selected pattern is the one that minimizes the Euclidean distance between the prediction and the positive patterns. This selected pattern is considered as a measurement for the correction step of the KF algorithm. The experiment results prove that the proposed method has the robust ability to track the moving human being across the image sequences with different challenging situations such as occlusion, deformation, rotation and the scale variation.