RECTGAUSS-Tex : block-based background subtraction

This paper presents an approach to background subtraction based on rectangular regions (blocks). The general principle is to successively divide the image into blocks and detect foreground pixels based on the color histogram and the variance between pixels of the blocks. Then, the classic Gaussian Mixture background subtraction method is applied to refine the detected foreground. Results show that this approach reduces false positives by filtering noise coming from small motion as it is based on groups of pixels instead of on individual pixels.