An improved segmentation algorithm to detect moving object in video sequences
Jinkui Li,Xinzhu Sang,Yongqiang Wang,Binbin Yan,Chongxiu Yu
DOI: https://doi.org/10.1117/12.869970
2010-01-01
Abstract:The segmentation of moving object in video sequences is attracting more and more attention because of its important role in various camera video applications, such as video surveillance, traffic monitoring, people tracking. and so on. Conventional segmentation algorithms can be divided into two classes. One class is based on spatial homogeneity, which results in the promising output. However, the computation is too complex and heavy to be unsuitable to real-time applications. The other class utilizes change detection as the segmentation standard to extract the moving object. Typical approaches include frame difference, background subtraction and optical flow. A novel algorithm based on adaptive symmetrical difference and background subtraction is proposed. Firstly, the moving object mask is detected through the adaptive symmetrical difference, and the contour of the mask is extracted. And then, the adaptive background subtraction is carried out in the acquired region to extract the accurate moving object. Morphological operation and shadow cancellation are adopted to refine the result. Experimental results show that the algorithm is robust and effective in improving the segmentation accuracy.