复杂交通场景中多运动目标分割算法.pdfVIP

2006年 3月 北 京 航 空 航 天 大 学 学 报 March摇 2006 摇 第32卷 第3期 Journal of Beijing University of Aeronautics and Astronautics Vol.32摇 No郾3 复杂交通场景中多运动目标分割算法 李俊韬摇 摇 张摇 海摇 摇 范跃祖 (北京航空航天大学 自动化科学与电气工程学院,北京 100083) 摇 摇 摘摇 摇 摇 要:针对室外复杂交通场景中多运动目标分割问题,提出了一种由变化检测和 运动分割组成的算法,利用水平集算法对帧差图像进行变化检测得到运动窗口,在运动窗口范 围内进行改进的k鄄均值聚类分割,利用运动相似性进行分割区域融合.算法避免了整个图像的 分割,减少了运算量,完整的分割出运动目标.试验结果表明,算法不仅能从复杂交通场景图像 序列中有效的检测和提取出运动目标并有很强的鲁棒性,而且能够解决运动目标的遮挡问题. 关摇 键摇 词: 目标检测;水平集;k鄄均值聚类;鲁棒性 中图分类号:TP391 文献标识码:A摇 摇 摇 摇 文 章 编 号:1001鄄5965(2006)03鄄0297鄄04 Multiple moving objects segmentationalgorithm incomplex traffic scene Li Juntao摇 Zhang Hai摇 Fan Yuezu (School of Automation Science and Electrical Engineering,Beijing University of Aeronautics and Astronautics,Beijing 100083,China) Abstract:To solve multiple moving objects segmentation problem in complex traffic scene outdoors,a new segmentation algorithm of multiple moving objects segmentation in complex traffic scene of outdoors was developed. The presented method consists of two phases. In the motion detection phase,motion鄄changed re鄄 gion was determined using an improved level set algorithm based on Mumford鄄Shah model of two consecutive frames difference of image sequence,and an simplified clustering algorithm providing initialization evolution curves of level set algorithm. Motion window was generated through detecting motion鄄changed region. In the motion segmentation phase,regions segmentation using an improved weighted k鄄means clustering algorithm of moving windows area were merged based on region motion similarity criterion. Computation time was reduced because of avoiding processing of thewhole image. Very promising experimental resultsindicatesthat thepro鄄 posed algorithm iseffectivewith strongrobustnessusingreal ima