A Novel Thresholding Approach to Background Subtraction
Lihao Jia, Yikun Liu · 2008
We propose a novel thresholding approach based on a new color space model and embed it into background subtraction. This model uses each pixel's color distortion and brightness distortion to detect the changes. The color distortion considers the vector's position in color space so that it can assemble the color features effectively. Moreover, this thresholding method also removes the moving shadow to some extent. By applying it to background subtraction, we get a comparative complete foreground object. For relatively complex background, we are motivated by the hysteresis threshold in Canny edge detection algorithm and introduce dual-threshold into background subtraction. It also achieves robust detection for videos containing complex background. We compare our method with other multi-model techniques, and the test results show the feasibility of the proposed algorithm.