A novel contour extraction method for motion object detection in surveillance video systems
Li Xie,Lingling Hu,Yiping Lü,Gang Xiong,Yaowu Chen
DOI: https://doi.org/10.3969/j.issn.1001-0505.2012.S1.007
2012-01-01
Abstract:Aiming at the problem that the traditional Gauss mixture model method has the shortcomings of being interfered easily and needing a large amount of calculations, this paper presents a contour extraction method used in intelligent video surveillance system for moving object detection. First this paper introduces the commonly used method of moving target detection; then it describes the traditional mixed Gauss model method, and analyzes the existed shortcomings in the target detection; it puts forward a new contour extraction method. This method uses pixel block-based Gaussian mixture model approach to extract the foreground object, uses mathematical morphology method to achieve foreground connectivity, using Freeman chain code to find the outline, uses Douglas-Peucker algorithm to fit the contour, and uses image moment to extract the centroid of target contour. Finally, experiments verify the proposed method, and it is compared with the traditional Gaussian mixture model approach. The experimental results demonstrate that the proposed method can remove the noise effectively and extract the object contour accurately.