一种改进的视频兴趣区目标捕获算法研究  

An Improved Target Capture Method of Video Interest Region

作　　者：周浩[1] 周平[1] 胡耀武[1] 

机构地区：[1]浙江理工大学信息学院,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2012年第4期565-569,共5页Journal of Zhejiang Sci-Tech University(Natural Sciences)

摘　　要：对于视频兴趣区的目标捕获技术提出了一种基于颜色梯度的背景差方法,通过结合阴影去除和连通域分析捕获兴趣目标。实验结果表明:该方法能有效解决光照突变等干扰,较好捕获兴趣区目标,减小误检率。对所实验的6个视频进行分析,在最多同时出现5个目标的视频检测中取得了92.8%的准确率、2%的误检率,且基本满足实时性。A new background difference algorithm for interest region in video is proposed, which is based on color gradient by means of integrating shadow removal and connected domain analysis to capture the interesting target. Experimental results demonstrate that the method is capable of dealing light sudden changes as well as capturing target＇s region of interest, in addition to its effectiveness for eliminating non- interest region＇s error detection. The experiment of 6 videos with 5 targets at most, the authors obtain in the high precision and low by mistake examining, Correctly 92. 8%, by mistake examining rate of 2%, and basically meet real-time processing.

关 键 词：目标捕获 背景建模 混合高斯模型 颜色梯度 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]