基于距离加权平均绝对差的模板漂移抑制算法

A template drift suppression algorithm based on distance weight MAD

针对图像序列目标匹配跟踪中出现模板漂移导致目标丢失的问题,提出一种基于距离加权平均绝对差的模板漂移抑制算法。对传统的最小绝对差准则进行改进,利用模板边缘到中心的距离作为参数对实时图与基准图的绝对差结果进行加权,增大失配位置漂移误差,使得真实位置的绝对差最小以防止模板更新过程中产生漂移。研究结果表明:该算法可以在跟踪中有效抑制模板漂移,实现对形变目标的长时间稳定跟踪,实时性好,便于在实时系统中实现。

Aimed at the problem of losing target caused by template drift in tracking with matching method, a template drift suppressing algorithm was put forward based on distance weight Minimum absolute difference （MAD）, which improves the traditional MAD principle. In the algorithm, the distance from template border to center was used as weighted parameter for calculating AD （Absolute difference） of real-time image and base image, which enlarged the drift error of mismatched position. The above process can minimize the AD of real position and prevent the template from drifting during template updating. The results show that this algorithm can suppress the template drift effectively in tracking and can track the deformable target stably for a long time. Furthermore, it owns good real-time performance and can be realized in real-time system conveniently.