摘要
针对背景差法易受外界环境因素影响的缺点,提出了一种基于改进K-均值聚类的背景建模方法。通过比较任意样本与该像素位置处的子类中心之间的距离,对各个像素的观察值进行聚类,并在聚类过程中逐步确定其类别数。一段时间的学习之后,样本数最多的子类就构成了背景模型。仿真结果表明,该算法即使在运动目标存在的情况下也能准确的提取出实际的背景,而且显著地降低了系统的存储量。
Aimed at the disadvantage that background subtraction was liable to be affected by outside environ-ment,a background modeling method based on the improved K-mean clustering was provided.By comparing the dis-tances between certain sample and sub-class center of the pixel,observation values of the pixels were clustered and the number of the clusters was determined during the clustering process..After learning for a period,background model was built by sub-class with the maximum samples.Simulations show that actual background can be extracted accurately by the algorithm even when moving targets are existent and the memory cost of the system is reduced dramatically.
出处
《电子测量与仪器学报》
CSCD
2010年第12期1114-1118,共5页
Journal of Electronic Measurement and Instrumentation
基金
甘肃省教育厅研究生导师基金项目(编号:0914ZTB003)资助项目
关键词
背景差
背景建模
K-均值聚类
背景更新
background subtraction
background modeling
K-mean clustering
background update
