基于视频的运动目标检测与识别  

Moving Object Detection and Recognition Based on Video

作　　者：张建军[1] 黄山[1] 张洪斌[1] 杨权[1] 汪勤[1] 

机构地区：[1]四川大学电气信息学院,四川成都610065

出　　处：《现代电子技术》2009年第8期87-91,共5页Modern Electronics Technique

基　　金：四川省重点技术创新项目计划资助项目(07XM020)

摘　　要：为了对目标进行跟踪和识别,针对固定场景提出一种基于视频的运动目标检测和识别算法。该方法采用改进的混合高斯模型为动态背景更新方法,并结合梯度背景差分方法及Otsu′s阈值分割法提取出运动目标,然后利用目标特征参数建立目标链,通过两条目标链间的目标特征匹配实现运动目标的快速跟踪与行为识别。该方法与传统方法相比具有更好的学习能力,提高了算法的场景适应性,从而有效地提高了运动目标检测的正确率和快速性。实验结果表明该方法具有良好的鲁棒性和自适应性。To track and recognize objects, an approach of detecting and recognizing moving objects in stationary scene is presented. A modified mixture Gaussian model is used as the dynamic background updating method. Moving objects are extracted by using gradient background difference method and Otsu＇s threshold segmentation and and then using objective parameters to establish the link of targets. Tracking targets and recognizing behaviors of targets by using a basic feature matching algorithm to targets of two links. Compared with the traditional method,it has better ability to learn, raise the scene adaptation of the algorithm,which has effectively improved the movement to the correct target detection rate and rapid. The results of experiment show that the method is robust and adaptive.

关 键 词：背景模型 混合高斯模型 运动目标检测与识别 Otsu’s阈值分割法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]