摘要
在图像多目标跟踪问题中,针对图像匹配无法辨别同类别目标以及状态滤波难以对目标突发机动建模两个难点,提出了一种多特征匹配融合跟踪算法。该算法在基于局部方差图(standard deviation map,STDM)的目标检测结果的基础上,首先利用目标感兴趣区域(region of interest,ROI)的图像匹配来克服目标状态匹配误差的影响;然后利用状态特征匹配消除图像匹配识别的模糊性;最后在关联代价全局最优化框架下,将两者匹配结果融合,以提高多目标跟踪的正确率。
There are two problems when tracking multiple targets in sequence images. Firstly, the target's kinematic state cannot be estimated accurately when target's abrupt maneuver happens. Secondly, the image matching method cannot discriminate the targets which belong to the same category. To resolve above problems, a novel muhi-target tracking method based on fusion of target's feature matching is proposed. On base of target detection in local standard deviation map (STDM), the region of interest's(ROI) image matching result is used to reduce the error of state estimation matching, and the state estimation matching is used to reduce the ambiguity of image matching. Under the global optimal association cost frame, fusion of above two matches is realized to improve the accuracy of tracking, which can resolve the complex multi-target tracking problem availably.
出处
《中国图象图形学报》
CSCD
北大核心
2008年第3期580-585,共6页
Journal of Image and Graphics
关键词
多目标跟踪
目标检测
关联代价
匹配
multi-target tracking, target detection, association cost, matching