基于鲁棒背景运动补偿的运动目标检测算法  

Moving Target Detection Algorithm Based on Robust Background Motion Compensation

作　　者：赖作镁[1] 王敬儒[1] 张启衡[1] 

机构地区：[1]中国科学院光电技术研究所国家"863"计划光束控制重点实验室

出　　处：《计算机应用研究》2007年第3期66-68,共3页Application Research of Computers

基　　金：国家"863"计划资助项目(2003AA823050)

摘　　要：提出一种用仿射参数模型来近似场景中摄像机的复杂运动,采用参数化的多分辨率估计方法鲁棒地估计出仿射参数;然后在当前帧与运动补偿后的帧之间求光流场,得到目标轮廓的初始分割;最后通过聚类和搜索填充算法分割出完整的目标。试验结果表明,该运动补偿算法能有效消除摄像机运动引起的背景运动,在摄像机运动情况下得到完整的目标。A new algorithm that affine model simulates cameras＇ complex motion in the scene was put forward, and then a precise motion compensation frame after robust estimation affine parameters through multi-scale motion estimation algorithm was obtained. After threshold the magnitude of optical flow between current frame and motion compensation frame, an initial target segmentation was obtained. At last, an intact target mask was detected through cluster and search fill algorithm. Experimental results show that this algorithm can eliminate global background motion very effectively because of cameras and segment target under moving cameras.

关 键 词：目标检测 背景运动补偿 多分辨率估计 光流 搜索填充 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]