An Efficient Object Segmentation Algorithm for Surveillance Systems

In this paper, we propose a novel method for detecting moving objects in a video sequence. The method is mostly suitable for video surveillance sequences in which background has no motion and changes are due to background changes (e.g. illumination changes and changes due to adding or removing parts of background) and the moving objects. In the first stage, we compute the difference image which is the difference between the background image and the coming image. The background image is obtained using a novel method. After that, we divide the difference image into blocks of equal size, and using mean and standard deviation of each block the difference image is divided into two regions at a coarse level (block level): foreground and background. To extract boundaries, we continue the procedure at the pixel level. Finally post processing is needed to eliminate false detection due to noise and eliminating shadow effects. Our proposed method has the ability of detecting multiple objects without knowing the number of objects a priori. In addition, a novel background update is proposed to cope with the changes of the background image.