多视频运动对象实时分割及跟踪技术  

Real-time segmentation and tracking technique of multiple moving objects in video sequence

作　　者：孙志海[1] 朱善安[1] 

机构地区：[1]浙江大学电气工程学院,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2008年第9期1631-1635,1660,共6页Journal of Zhejiang University：Engineering Science

摘　　要：利用帧间差异积累信息进行自适应背景建模,采用背景差的方法检测视频运动对象区域.设计了一种变系数的空域滤波器,有效地对背景差图像进行了增强,使获得的视频运动对象区域更具有空域连通特性.给出了一种改进的基于Otsu法的自适应阈值化方法,能更准确地对背景差图像进行阈值化.采用形态学边界提取技术对视频运动对象轮廓进行提取.在获得视频运动对象轮廓的基础上,用区域生长法对视频运动对象进行定位,将矩形中心坐标视为视频运动对象的质心坐标.用基于空间欧氏距离最短的方法对每个视频运动对象质心进行关联跟踪并绘制轨迹.试验结果表明,该方法实时有效.Effective background was modeled by accumulative frame-to-frame differences, and foreground could be extracted using background subtraction operation. A variable coefficient spatial filter was presented, which could effectively enhance the background subtraction image and make the foreground region has much more spatial connectivity. An improved Otsu-based segmentation method was presented to accurately threshold the background difference image. Morphological operation was used in extracting contours of moving objects. After getting the contours of moving objects, region growing segmentation method was applied to locate the foreground region, and the rectangle centre was seen as the mass centre of the moving objects. Moving objects were tracked using minimal Euclidean distance and the motion trajectories were also painted. Experimental results showed that proposed algorithm was effective and efficient.

关 键 词：运动对象分割 运动对象跟踪 差异积累 空域滤波 背景更新 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]