Boundary and shadow position-based moving objects detection

Video moving objects detection has been an important task in video surveillance. As a single detection algorithm often results in detection error, especially when the characteristic of the object is similar with the background. In this paper, a new detection method based on boundary and shadow position is proposed to accurately detect object. Firstly, the pre-detection objects are extracted by using Sequential Kernel Density Approximation (SKDA) and RGB color space respectively. Then two boundary sets can be obtained by applying Sobel edge operator to the pre-detected objects, and the inconsistent boundary will be identified afterwards. For each inconsistent boundary pair, its associated curvature and shadow mark vector is used as criteria to evaluate the most probable location of the true boundary. The final object is extracted by choosing the merged pre-detection objects in region surrounded by the evaluated boundary. Experimental results show that the proposed algorithm performs better under different scenes compared with other existing algorithms.