铁路客运站复杂环境中的运动目标检测  

Moving target detection in complex environment of railway station

作　　者：孙首群[1] 刘康亚[1] 刘硕妍[2] 吕晓军[2] 詹璇[2] 

机构地区：[1]上海理工大学机械工程学院,上海200093 [2]中国铁道科学研究院电子计算技术研究所,北京100081

出　　处：《交通运输工程学报》2013年第3期113-120,共8页Journal of Traffic and Transportation Engineering

基　　金：国家863计划项目(2009AA11Z211);铁道部专项资金支持课题研究项目(J2011X007);上海市教委重点学科建设项目(J50503)

摘　　要：采用分层组织的形式将传统高斯混合模型分为背景层、竞争层和噪声层,各层分别采用不同的更新机制,在各层之间引入晋级和降级机制以纠正可能存在的误判。采用基于轮廓检测的噪声滤波实现噪声层更新以消除噪声,并利用直方图匹配检测伪前景区域以提高对背景变化的适应能力。使用停车场视频和铁路客运站候车室视频对改进后高斯混合模型的检测效果进行了验证。验证结果表明:改进的高斯混合模型有效避免了长期静止的目标被融入背景,降低了光线突变或摄像机噪声的干扰,加快了背景改变时模型的更新速度,目标检测速度比传统GMM提高了10%。检则方法满足了铁路客运站智能视频监控实时性和准确性的要求,为视频分析奠定了基础。Traditional GMM （Gaussian mixture model） was dived into background layer, completion layer and noise layer by using hierarchical organization. Diverse update mechanisms were applied in different layers. In order to correct possible misjudgment, promotion and downgraded mechanisms were introduced between layers. To eliminate noise, noise layer was updated by using noise filter based on contour detection. In order to improve the adaptability for changing background, pseudo foreground area was detected by using histogram matching. The detection effect of improved GMM was verified by using the videos of station and parking lot. Verification result indicates that the problem of long-term static target being merged into background is settled. The impact of light mutations or camera noise is reduced. The updating speed of model increases when the background changes. Detection speed increases by 10% compared with traditional GMM. The efficiency and accuracy of moving target detection in railway station are improved by improved GMM, and the foundation for intelligent video analysis is laid. 1 tab, 10 figs, 18 refs.

关 键 词：运动目标检测 高斯混合模型 直方图匹配 噪声滤波 分层组织 轮廓检测 

分 类 号：U491.116[交通运输工程—交通运输规划与管理]