LOBS: Local Background Subtracter for Video Surveillance

Abstract

In surveillance system video sequences are obtained through static cameras and fixed background. A popular approach called background subtraction is generally used in this scenario. Existing approaches in this field try to detect the object first and then remove the shadow in the subsequent phase. Here we have tried to combine both object detection and shadow removal module to a single module. In this work, a background model is proposed based upon the stationary pixels across the frames required for background model initialization. Considering the stationary and non-stationary pixel information background model is developed, which is used for background subtraction in subsequent phase. A local thresholding based background subtraction technique is proposed for foreground object extraction and removal of shadow. Experimental results shows that our method outperforms many state-of-the-art techniques. The proposed technique is robust to challenges like pose and illumination variations.