Human Cast Shadow Elimination Method Based on Orientation Information Measures

Abstract

Shadow elimination is critical for robust and reliable visual surveillance system. Because cast shadows have most motion features as the moving objects, it is difficult to eliminate it. This paper presents a novel moving cast shadow elimination algorithm. The algorithm is based on heuristic approach to detect and eliminate the human cast shadow. Moving region geometrical analysis approaches is proposed. Based on human body geometrical properties, the shadow existence and its approximate location can be determined. Shadow detection algorithm based on image orientation information measure is proposed. The frame ratio image is segmented into shadow-like region and non-shadow-like region. Finally, the approximate shadow location and the shadow-like region are fused to form the fine shadow region, and then the shadow can be eliminated. The algorithm has high efficiency and anti-noise performance. The experiments demonstrate shadow region can be eliminated quite well and thus good human body segmentation results can be obtained.