A Robust Background Subtraction Approach with a Moving Camera

In this paper, an algorithm for extracting moving objects from a moving background scene that contains complex camera motion is proposed. The robust and efficiently background subtraction algorithm that is able to cope with local scene changes, such as audiences and trees, as well as globe scene changes caused by camera motion. The algorithm is based on a proposed background subtraction method and comprises three steps: the feature-based camera motion estimation, panorama reconstruction and background subtraction. The background subtraction in our method, which introduces risk maps to account for the problem that the background image may not be perfectly aligned to the input images, greatly improves the foreground separation even on the object boundary. We have applied this method to standard image sequences. Experimental results show the proposed method can be successfully provided with the background subtraction capability.