基于证据理论和多核函数融合的目标跟踪  

Target tracking method based on evidence theory and multiple kernel function

作　　者：杨欣[1] 沈雷[1] 费树岷[2] 周大可[1] 

机构地区：[1]南京航空航天大学自动化学院,南京210016 [2]东南大学自动化学院,南京210096

出　　处：《东南大学学报（自然科学版）》2015年第5期861-864,共4页Journal of Southeast University：Natural Science Edition

基　　金：国家自然科学基金资助项目(61573182;61305011);光电控制技术重点实验室和航空科学基金联合资助项目(20145152027);南京航空航天大学青年科技创新基金资助项目(NS2014035);南京航空航天大学研究生创新基地(实验室)开放基金资助项目(kfjj201426)

摘　　要：针对目标长时间大面积被物体遮挡所导致的跟踪位置偏差或丢失,提出一种基于证据理论和多核函数融合的目标跟踪方法.首先,采用Mean-Shift方法分别计算边缘方向直方图模型和颜色直方图模型;其次,使用特征融合的方法将2种特征模型进行融合;再采用多核函数计算出多个相对应的目标位置估计;运用基于证据理论的方法对已获得的不同目标位置估计进行判断取舍和计算,计算出最理想的目标区域作为最终的目标位置.经过这样的处理和计算可以保证在遮挡的情况下能够实时准确地跟踪上原始目标.实验结果表明,所提算法在处理目标大面积遮挡、光照变化等问题时具有更好的性能和鲁棒性.When an object is occluded for long the tracking position deviation or loss may occur.To solve this problem a target tracking method based on evidence theory and multiple kernel function is proposed.First,edge direction histogram model and color histogram model are calculated by using Mean-Shift method,and then the two feature models are fused together by the feature fusion method.Secondly,multiple kernel function is adopted to obtain several estimated positions of target. Thirdly,according to evidence theory,judgment and calculation are implemented to get the final target location through the obtained multiple target position estimates.By these processing and calculation,it can be ensured that the original target can be tracked in real time and accurately in the case of occlusion.Experimental results show that the proposed algorithm has better performance and robustness for resolving the problem of occlusion and illumination change during target tracking.

关 键 词：目标遮挡 目标跟踪 证据理论 均值漂移 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]