扩展目标跟踪中的自适应模板刷新算法研究

Adaptive Template Update Algorithm Research for Extended Object Tracking

在目标跟踪领域基于模板相关匹配的跟踪算法是一种非常实用而且有效的方法,但是由于目标表面在运动过程中经常会发生变形,因此需要使用模板刷新的方法来适应目标的变形。而普通的模板刷新策略又会导致模板漂移,为了减少这种刷新策略带来的漂移,对于扩展目标,利用其拥有较多细节的特点,提出了一种基于尺度不变特征变特征点的方法,将发生仿射变换的模板与原始模板的比较,利用比较的结果来决策刷新模板的时机,实验表明该方法对于扩展目标的跟踪来讲,该方法能够很有效而合理刷新模板,减少模板漂移,提高了跟踪精度。

Template matching tracking algorithm is a kind of very practical and effective method in object tracking field, but due to the deformation of the object, template update has to be used for the deformation of the object, and ordinary template update method usually leads to template drifting. In order to avoid such drifting, considering the character of extended object containing more details, an adaptive template update method is proposed. It is based on SIFT （Scale Invariant Feature Transform） key points by which the affine transformation object template can be matched with original object template, and then the template can be updated. Experiments prove that the algorithm can relieve template drifting effectively and track the extended object better.