基于感兴趣区域的运动分割算法  

Motion Segmentation Method Based on the Region of Interest

作　　者：王锦[1] 于俊清[1] 段萌远[1] 

机构地区：[1]华中科技大学计算机科学与技术学院数字媒体研究室,武汉430074

出　　处：《微处理机》2008年第3期139-141,共3页Microprocessors

基　　金：华为基金(YJCB20050241N);湖北省自然科学基金(2005ABA265)

摘　　要：提出了一种基于感兴趣区域ROI(Region Of Interests)进行形态学处理的运动分割算法。首先对变化检测得到的二值图像进行分割(粗分割),然后对粗分割后的每个区域内部进行形态学处理,最后再次对经过处理后的二值图像进行分割(细分割)。该算法能够很好地处理噪声较大情况下的运动分割,避免了大面积噪声对于前景目标分割的影响。This paper presents a new method to get these regions of interests from the foreground pixels ,which makes use of contours and local areas information. Foreground objects are segmented from the background in each frames of the video sequence by a three stage process： roughly segmenting from those change - detected pixels, separately morphological processing with the results of the previous step, globally segmenting again from results of the last step, which similar with the roughly segmentation. In the case of large areas of noisy, the method can better still avoid the influence of them in items of motion segmentation, compared to the common method which with the usage of global morphological process.

关 键 词：变化检测 背景模型 帧间差分 运动分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]