基于前景策略和场景信息的目标跟踪系统  

Object tracking system based on foreground abstracting and scene information

作　　者：窦琴[1] 刘清[1] 郭建明[1] 

机构地区：[1]武汉理工大学自动化学院,武汉430063

出　　处：《大连海事大学学报》2009年第2期84-88,共5页Journal of Dalian Maritime University

基　　金：高等学校博士学科点专项科研基金资助项目(20060497017)

摘　　要：为解决目标跟踪算法中的某些难点问题,提出以Mean-shift算法为基础,基于高斯混合模型(GMM)前景分割和场景信息的MGSI方法。该方法基于运动预测和前景分割为目标跟踪提供感兴趣区域(ROI),解决了跟踪目标与背景相似情况下目标识别中的误报问题。同时,通过场景信息的预先设定来获取某些先验知识,如屏蔽区域的划分、区域中目标模板更新的频率、遮挡类型的预判等,并根据先验知识来调整跟踪策略,一定程度上解决了遮挡问题。实验证明,基于MGSI方法的目标跟踪系统在一定程度上解决了光照变化、背景干扰、模板更新、遮挡等传统跟踪系统中存在的问题,有效提高了跟踪的准确率。To solve some difficult problems in object tracking algorithm, an Mean-shift based on Gaussian mixture models （GMM） and scene information （MGSI） method was developed. The MGSI method provided tracking region of interest （ROI） based on the predicted position and foreground abstracted by GMM, which solved the false alarm in object recognition when the tracked object was similar with background. By setting information of the scene in advance, such knowledge as the differentiation of the shield region, object model updating frequency and shield type, etc. could be obtained. According to the knowledge obtained the tracking strategy could be adjusted and then the shelter problems could also be solved to a certain extent. Experimental results show that MGSI based on the tracking system has better performance in the cases of illumination changing, background disturbing and shelter, and tracking accuracy is improved effectively.

关 键 词：目标跟踪系统 MEAN-SHIFT算法 前景提取 场景信息 感兴趣区域(ROI) 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]