尺度方向自适应的减法聚类视频运动目标定位  

Scale and Direction Adaptive Locating of Video Moving Objects with Subtractive Clustering

作　　者：孙志海[1,2] 孔万增[1] 朱善安[2] 

机构地区：[1]杭州电子科技大学计算机学院,杭州310037 [2]浙江大学电气工程学院,杭州310027

出　　处：《光电工程》2010年第1期37-42,共6页Opto-Electronic Engineering

基　　金：浙江省重大科技专项(优先主题工业项目)资助项目(2008C11108-1;2008C13076);浙江省自然科学基金项目(Y1080883)

摘　　要：针对减法聚类算法对视频运动目标进行定位时无法获取目标尺度及方向参数的问题,本文提出了一种可获取待定位目标尺度及方向参数的视频运动目标定位算法。该算法在减法聚类算法预定位目标位置及获得目标个数的基础上,进一步采用模糊C均值聚类对目标前景样本进行归类,最后通过对目标前景样本协方差矩阵特征值和特征向量的分析获得目标的尺度及方向参数,从而实现对视频运动目标的定位。实验结果表明,所提出的方法与原减法聚类定位方法相比可获得更合理的目标定位结果。To overcome the problem that the common subtractive clustering object locating method could not acquire video object＇s scale and orientation parameters,an improved subtractive clustering object locating algorithm,that is called the scale and direction adaptive locating of video moving objects with subtractive clustering algorithm,is proposed. The proposed method first uses common subtractive clustering locating method to acquire video object position and moving object number of each frame. Then,the proposed algorithm uses fuzzy C-means clustering algorithm to further cluster the moving object foreground pixel samples. Finally,matrix analysis theory for calculating eigenvalues and eigenvectors of covariance matrix is applied to compute the object scale and orientation parameters. Experiment results show that the proposed object locating algorithm could obtain much more reasonable video object locating results.

关 键 词：减法聚类 目标定位 目标检测 模糊C均值聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]