基于高斯混合模型的铁路入侵物体目标识别方法  

Target Recognition Method of Railway Invasion Based on Gaussian Mixture Model

作　　者：董宏辉[1] 孙智源[1] 葛大伟[1] 秦勇[1] 贾利民[1] 

机构地区：[1]北京交通大学轨道交通控制与安全国家重点实验室,北京100044

出　　处：《中国铁道科学》2011年第2期131-135,共5页China Railway Science

基　　金：国家"八六三"计划项目(2006AA11Z231);铁道部科技研究开发计划项目(2007X007-A);北京市科技计划重点项目(D07020601400707)

摘　　要：基于铁路的特殊视频场景,研究铁路入侵物体目标识别技术,提出并应用1种改进的高斯混合模型,拟定合适的颜色变化阈值和背景更新速率,对图像的不同部分分别进行背景更新,实现铁路视频的背景建模,并由此得到稳定的背景图像;通过背景像素与前景像素的贝叶斯分类实现对铁路入侵物体的准确检测。对典型的铁路入侵行为视频进行实验分析,结果表明:应用改进的高斯混合模型可以更好地适应场景的变化,并能够更加快速、准确地实现在铁路环境下对入侵物体的目标识别。Based on the specific video of railway scene, the target recognition of railway invasion is studied, and an improved Gaussian Mixture Model （GMM） is proposed in this paper. This improved GMM can get more stable backgrounds, by setting the different values of the color changing threshold and background updating rate, and by updating the different image parts, respectively. Using the Bayesian classification for the background pixel and foreground pixel, the railway invasion is detected accurately. The experiment results on the typical railway video show that the improved GMM can adapt to the scene changing better, and make the invasion target recognition more quickly and accurately in the railway conditions.

关 键 词：高斯混合模型 背景建模 运动检测 目标识别 铁路安全 

分 类 号：U298.12[交通运输工程—交通运输规划与管理]