基于区域颜色聚类的运动目标阴影检测  

Moving Object Shadow Detection Based on Regional Color Clustering

作　　者：徐杰[1] 项文波[1] 茅耀斌[1] 孙金生[1] 

机构地区：[1]南京理工大学自动化系,江苏南京210094

出　　处：《计算机技术与发展》2016年第3期193-196,共4页Computer Technology and Development

基　　金：国家科技重大专项基金(2011ZX04002-051)

摘　　要：文中提出一种基于区域颜色特征的阴影检测方法。首先,通过均衡化灰度直方图来增强图像中各个区域之间的对比度;使用K-means算法对前景区域进行聚类。对于聚类后的每个区域,统计每部分的色彩特征不变量直方图,利用一种直方图相似性度量方法来比较当前区域和对应背景区域的相似性,根据相似性结果对每个区域进行阴影属性的判断。通过与基于彩色特征不变量、基于HSV颜色空间等方法进行比较,结果表明文中算法的阴影检测率和前景检测率有一定的提升,且用时较少。实验结果表明,文中算法充分利用了阴影的区域颜色特征,在多类场景中能够有效检测出阴影,且具有较好的实时性。A novel algorithm based on regional color feature is proposed to detect moving shadow. First,an image equivalence method is built to enhance the contrast between regions in the frame, and K- means is also used for the foreground cluster. Then,the statistical histogram of color invariant feature is calculated in each cluster,comparing the similarity of histogram between foreground and background by histogram similarity measuring method. Last,the shadow properties of regions are judged according to the similarity result. By comparison with certain classical methods such as invariant color features based method, HSV color space based method and so on, the proposed method performs better than some of them in terms of shadow detection rote and running time. The experimental results show that the al- gorithm in this paper make full use of the color feature of region to detect the shadow effectively in multi-class view and has better real -time.

关 键 词：视频监控 运动目标阴影检测 增强对比度 彩色特征不变量 K-MEANS 直方图相似性 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]