摘要： 跟踪窗宽固定的目标跟踪算法,由于目标尺度变化造成窗口内背景信息量增加或目标信息量减少,可能导致跟踪失败。针对该问题,提出一种基于碎片表征的尺度自适应目标跟踪算法。利用结构化输出支持向量机模型确定目标的空间位置,将目标图像区域分割成多尺度矩形碎片,通过地表移动距离度量相似度,在尺度空间搜索与模板碎片最相似的目标碎片,确定目标特征尺度。实验结果表明,该算法不仅可适应目标尺度变化,而且在目标姿态变化、亮度改变和局部遮挡的情况下,具有准确稳定的跟踪性能。与LOT,OAB和CXT算法相比,平均跟踪误差较低,跟踪性能较好。

关键词: 目标跟踪, 尺度自适应跟踪, 碎片表征, 结构化输出支持向量机, 地表移动距离

Abstract: Object tracking algorithm with fixed tracking window may cause failure due to the increase of background information in tracking window for a smaller target or the decrease for a larger target.To solve this problem,a scale-adaptive target tracking algorithm based on fragments representation is proposed.Firstly,structured output Support Vector Machine(SVM) is used for spatial locatlization,then the object is represented by multiple image fragments.Finally,fragments of different scales are compared with Earth Mover’s Distance(EMD) metric to determine the final scale.Experimental results show that the proposed algorithm cannot only cope with scale variation but behave well in appearance change,brightness change and partial occlusion of target.The tracking error is decreased which has a better performance compared with existing algorithms including LOT,OAB and CXT.

Key words: target tracking, scale-adaptive tracking, fragments representation, structured output Support Vector Machine(SVM), Earth Mover’s Distance(EMD)

中图分类号: 

• TP391