基于FAST特征点改进的TLD目标跟踪算法

Improved TLD Object Tracking Algorithm Based on FAST Feature Points

TLD是近年来出现的一种较为新颖的长时间目标跟踪算法,它与传统的跟踪算法区别在于将传统的检测算法和跟踪算法结合起来,通过学习模块来学习目标的特征,从而进行有效跟踪.笔者针对算法跟踪器无法可靠跟踪均匀选取的特征点的问题,提出一种基于FAST特征点改进的TLD目标跟踪算法,保证所选特征点能够被正确可靠跟踪,提高跟踪器的精度.针对跟踪过程中学习模块的模板累积效应明显,实时性降低的问题,采用一种动态模板管理机制.在模板数量达到阈值时,通过比较模板与当前目标的相似度,删除特定模板,保持模板数量恒定.实验表明,改进后的算法具有更高的跟踪精度和实时性.

TLD was a new long-term object tracking algorithm in recent years.It is different from the traditional tracking algorithm in that it combined the traditional detection algorithm and the tracking algorithm,and then the learning module was used to study the characteristics of the object.In this paper,an improved TLD object tracking algorithm based on FAST feature points was proposed to ensure that the selected feature points could be tracked correctly and reliably to improve the accuracy of the tracker.At the same time,for the tracking process,the learning module template cumulative effects is obvious,reduced the real-time performance.Using a dynamic template management mechanism,when the number of templates reached the threshold,by comparing the similarity between the template and the current target,the specific template was deleted and the number of templates was kept constant.Experiments results showed that the improved algorithm had higher tracking precision and real-time performance.