煤矿智能视频监控中的运动目标检测研究  

Research on moving target detection in coal mine intelligent video monitoring

作　　者：张谢华[1] 赵小虎[2] 

机构地区：[1]江苏师范大学智能教育学院,江苏徐州221116 [2]中国矿业大学物联网(感知矿山)研究中心,江苏徐州221008

出　　处：《工矿自动化》2016年第4期31-36,共6页Journal Of Mine Automation

基　　金：国家科技支撑计划项目(2012BAH12B00);江苏师范大学博士学位教师科研支持项目(2015XLR18)

摘　　要：针对煤矿智能视频监控环境存在各种复杂动态场景变化的情况,研究了运动目标检测中的3个重要环节:背景建模与更新、前景检测和运动阴影检测与去除。针对这3个环节,提出了相应的处理方法:基于IFCM聚类算法的自适应背景建模与更新方法,对像素灰度取值进行无监督聚类,自适应选取不同个数的聚类构建各像素背景模型,随场景变化进行聚类修改、添加和删除以完成背景自动更新;联合背景差分信息、三帧差分信息和空间邻域信息的前景检测方法,据此获得较为准确的前景目标;运动阴影检测与去除方法,依据在阴影覆盖前后的灰度图像中,像素具有亮度值相关性和纹理特征值不变性,实现了运动阴影的检测与去除。实验结果验证了本文所提方法的有效性和优越性。In view of condition of complex dynamic scene changes in coal mine intelligent video monitoring environment,three important steps in moving target detection were researched which were background modeling and updating,foreground detecting,motion shadow detecting and removing.For the three steps,corresponding processing methods were put forward：a self-adaptive background modeling and updating method based on IFCM clustering algorithm was proposed,the method was used to unsupervised clustering of pixels＇ gray values,different number of clusters was adaptively selected to construct the pixels＇ background model,and automatic updating of the background model was completed by modifying,adding and deleting clusters with the scenes＇ change;a foreground detection method was proposed which combined the background difference,three frame difference and spatial neighborhood information,so accurate foreground targets were obtained;a motion shadow detection and removal method was proposed on the basis that the pixels＇ of gray images have characteristics of luminance correlation and texture invariance before and after shadow covering,so detecting and removing moving shadow was realized.The experimental results verify effectiveness and superiority of the method.

关 键 词：煤矿视频监控 运动目标检测 背景建模 前景检测 运动阴影检测 

分 类 号：TD67[矿业工程—矿山机电]