基于像素值聚类的运动目标检测方法  

Moving target detection method based on pixel clustering

作　　者：蒋建国[1] 郎立娜[1] 齐美彬[1] 安红新[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《电子测量与仪器学报》2012年第2期101-106,共6页Journal of Electronic Measurement and Instrumentation

基　　金：中央高校基本科研业务费专项资金(2011HGZL0001)

摘　　要：提出了一种基于像素值聚类的运动目标检测方法。在RGB空间进行背景提取,将一幅图像中的每个像素点在时间轴上进行聚类,提取背景模板;之后采用当前图像与背景模板进行比较,从而检测出运动目标,并同时进行背景模板的更新。在此过程中采用了模板替换机制以节约存储空间;在背景建模部分加入结束判定部分,以自适应的建立背景模板。相比于经典的混合高斯(GMM)和codebook方法,该算法的运行时间至少缩短了1/4,提高了检测速度。实验结果表明,该算法能够在不同条件下快速准确的检测运动目标。A moving target detection method based on pixel clustering is proposed in this paper. The method is based on clustering for each pixel on a timeline in RGB space. The initial backgrounds are extracted from surveillance videos. Then the current image and the background templates are compared and the objects are detected correctly. And background templates are renewed at the same time. Replacement of the templates is used to save memory space. And judgment at the end of modeling is used to extract the background adaptively. Compared with the mixed Gaussian method and the codebook method, the proposed method at least saves a quarter of time and improves the detection rate. The experiments demonstrate that the method can effectively and correctly detect objects under various conditions.

关 键 词：像素值聚类 背景建模 运动目标检测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]