一种自适应阈值的运动目标提取算法  

Adaptive threshold algorithm for extracting moving objects

作　　者：孙明玉[1] 丁莹[1,2] 李文辉[1] 江琦[1] 徐长青[1] 

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]长春理工大学计算机科学技术学院,长春130022

出　　处：《计算机应用研究》2010年第6期2380-2383,共4页Application Research of Computers

基　　金：国家"863"计划资助项目(2008AA10Z224);国家自然科学基金资助项目(60873147;60573182);国家教育部博士点基金资助项目(20060183042);吉林省科技发展计划资助项目(20060527)

摘　　要：为了准确地划分运动目标和背景区域,提出一种自适应阈值的运动目标提取算法,对现有基于背景差的提取算法进行改进。本算法将运动目标和背景作为两个聚类,对图像中的点按像素灰度进行分类,以聚类间的方根—算术均值距离最大作为分割阈值选择的准则,使得运动目标提取算法中二值化阈值能够自动更新,从而实现对运动目标的准确完整提取。实验结果表明,该算法能够较准确快速地提取运动目标,并对环境亮度突变、背景存在微小运动等情况具有较好的鲁棒性。This paper presented a moving objects extraction algorithm based on adaptive threshold, which had improved extraction algorithm based on background subtraction. Moving objects and background were regarded as two clusters, and classified the intensity values at the pixels, then adopted the max clusters’ square root arithmetic mean divergence as the threshold selection rule. So the threshold could be automatic update, in order to accomplish exactly and perfectly extract. Simulation results indicate that motion objects can be extract correctly and fast by using the new algorithm, even in the case of dynamic natural environments since they include motions like swaying vegetation, breaking luminance etc.

关 键 词：运动目标提取 背景差 阈值 方根—算术均值距离 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]