Moving human detection in video using dynamic visual attention model

Abstract

Visual Attention algorithms have been extensively used for object detection in images. However, the use of these algorithms for video analysis has been less explored. Many of the techniques proposed, though accurate and robust, still require a huge amount of time for processing large sized video data. Thus this paper introduces a fast and computationally inexpensive technique for detecting regions corresponding to moving humans in surveillance videos. It is based on the dynamic saliency model and is robust to noise and illumination variation. Results indicate successful extraction of moving human regions with minimum noise, and faster performance in comparison to other models. The model works best in sparsely crowded scenarios.