基于多特征组合的协方差目标跟踪方法  

Covariance tracking based on multi-characteristic combination

作　　者：刘清[1] 窦琴[1] 郭建明[1] 李龙利[1] 

机构地区：[1]武汉理工大学自动化学院,湖北武汉430063

出　　处：《武汉大学学报（工学版）》2009年第4期512-515,共4页Engineering Journal of Wuhan University

基　　金：高等学校博士学科点专项科研基金(编号:20060497017)

摘　　要：基于多特征组合的协方差矩阵表征目标的方法,研究用协方差矩阵来描述跟踪的感兴趣区域(ROI),从而提出基于速度预测和前景的协方差目标跟踪方法.因为积分图方法可以加快协方差的计算速度,在提出的协方差跟踪方法中使用了积分图快速算法,进一步提高算法的效率.结合目标速度的预测和前景提取缩小搜索范围,加快了匹配速度,使所提出的协方差目标跟踪方法能更进一步提高跟踪效率和准确性.通过背景强干扰、光照变化和相同颜色遮挡情况下的目标跟踪实验,结果表明基于速度预测和前景的协方差的跟踪方法在复杂场景下跟踪的准确性很高,跟踪的鲁棒性和快速性有明显提高.The covariance matrix fused by multiple features is used as the object descriptor and denoting the region of interest （ROI） of tracing. The covariance tracking in video sequences is proposed based on position predicting and foreground segmentation. By using integral image the calculation speed of covarianee matrix is accelerated. And object speed prediction can reduce the scope of the region of object matching. Object tracking efficiency is periments of object tracking raised. Foreground segmentation made the tracking more accurately. The exin video sequences in the presence of cluttered backgrounds and sudden illumination changes and shelter are finished. Experimental results indicate that the tracking way presented has good performances under complex background and the tracking robustness and speed are enhanced greatly.

关 键 词：多特征组合 协方差 目标跟踪 积分直方图 前景提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]