Background subtraction using self-identifying patterns

Separation of object foreground from background is used in 3D model creation and matting in video production. Robust background subtraction techniques that function in uncontrolled lighting environments would be useful for many applications. We introduce a method using bi-tonal self-identifying patterns as a background that can be used to recognize the foreground object despite the background intensity and colour being non-uniform across the image. Detected pattern points are used to sample the black and white colour levels in several image points. A surface is fitted to both the black and white colour levels allowing an estimated background image to be created. The background image is then subtracted from the original image to isolate the foreground objects. The method of using self-identifying patterns also provides the camera-pattern pose for use in 3D model creation. A visual hull 3D model can be created by identifying the outline of an object from several known camera poses. Examples of this method applied to both matting and 3D model creation are given. Experimental results are shown.