基于差异积累的视频运动对象自动分割  

Automatic segmentation of moving objects in video sequence based on difference accumulation

作　　者：孙志海[1] 朱善安[1] 

机构地区：[1]浙江大学电气工程学院,浙江杭州310027

出　　处：《光电工程》2007年第12期97-103,共7页Opto-Electronic Engineering

摘　　要：针对视频运动对象的自动分割,本文给出了一种基于差异积累的自动分割算法。与传统的基于运动信息变化检测方法不同,该算法通过累积的帧差信息构建出可靠的背景,与当前帧比较进而提取出视频运动对象。本文提出了一种增强的基于Otsu法的自适应阈值化方法,能更准确地对背景差图像进行阈值化分割,克服了传统Otsu法阈值化容易失效的问题。改进的基于区域生长的定位方法更能避免传统方法的误定位及重定位的问题。实验结果表明,本文算法具有较好的实时性、自适应性以及鲁棒性,可以较为可靠地建立背景模型并进行实时更新,适用于刚体或非刚体存在平缓的光照变化以及摄像头微抖动的视频运动对象的自动分割。An automatic segmentation method is presented for moving objects in video sequence based on differences accumulation information. Effective background is modeled by accumulative frame-to-frame differences, and moving objects in video sequence could be extracted using background difference, which is different from conventional movement-based detection methods. Enhanced Otsu segmentation method is presented, which segments the background difference image much more accurately, and overcomes conventional shortcomings of Otsu method. Improved localization method based on region growing could avoid repeated location and false location. Experiments are implemented by using different video sequences under different circumstances. The results show that proposed algorithm is efficient, adaptive, robust, and could handle various scenes, including rigid and non-rigid object segmentation, smooth illumination changes, and camera dithering

关 键 词：差异积累 最大类间方差法 背景更新 视频对象分割 区域生长 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]