一种基于场景区分算法的背景抽取方法  

A background extraction method based on scene distinction algorithm

作　　者：杜向龙[1] 伍健荣[1] 邢涛[1,2] 

机构地区：[1]中国科学院上海微系统与信息技术研究所无线传感器网络与通信重点实验室,上海200050 [2]无锡物联网产业研究院,江苏无锡214135

出　　处：《传感器与微系统》2011年第11期39-42,共4页Transducer and Microsystem Technologies

基　　金：国家科技重大专项基金资助项目(2010ZX03006-004);国家重点基础研究发展计划资助项目(2011CB302906)

摘　　要：视频监控面对的场景具有静态的成分和缓变的成分。为了从视频流中稳定地抽取较精确的背景,提出了一种基于场景区分的背景抽取算法。首先通过混合高斯模型对视频序列中稳定的成分进行建模,然后采用Kalman滤波模型对视频序列中变化的成分进行建模,最后采用场景的区分算法将2种模型结果进行处理从而得到背景。实验结果表明:该方法能从视频流中稳定地抽取相对精确的背景。The scene monitored by intelligent video is stable most of the time and slow changing temporarily. In order to stably and precisely extract the background from the video stream, a background extract algorithm based on the scene distinction has been presented. The mixture of Gaussians and Kalman filter are employed to process and model the stationary parts and slow changing parts separately, and then the scene discrimination algorithm is used to extract the background by processing the result of those two models. Experimental results indicate that the method can stably extract relative precise background from the video stream.

关 键 词：背景抽取 混合高斯模型 卡尔曼滤波器 场景区分 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]