ABSTRACT
The detection of regions of interest is the important step in many computer vision applications, such as visual surveillance, event detection, image understanding etc. A general object detection technique is desired, but it is very dicult to detect objects even unknown objects under complex environments. So many computer vision applications assume stable environments or few changes in scenes, in which the background model is trained and foreground regions are obtained via the dierence between current frame and the background model, the task of object detection implementation is straightforward. The process is called background subtraction.
