基于改进的TLD目标跟踪算法

针对传统跟踪—学习—检测(tracking-learning-detecting，TLD）目标跟踪算法由于检测模块扫描大量子窗口而导致检测时间过长，并且在跟踪过程中当目标发生严重遮挡、形变时，TLD算法会出现跟踪失败的问题进行了研究，提出改进TLD目标跟踪算法。改进算法在检测模块前加入ViBe模型预估前景目标，极大地缩小了检测区域。追踪模块用SIFT特征匹配算法来代替原算法中的光流法，准确跟踪目标避免发生跟踪漂移，减少了计算的复杂度，提高了算法适应环境的能力。实验表明，改进后的TLD算法运行速度得到提升，并且当目标出现严重遮挡、光照强度剧烈变化时的跟踪精度也得到了很好的改善。

Target tracking algorithm based on improved TLD

Aiming at the problems of that the detecting module scans a large number of sub windows, which result the detection time is too long , and when the target has severious occlusion and deformation during the tracking process, the traditional tracking learning detection (TLD) target tracking algorithm will fail to track, so this paper propose the improved TLD target tracking algorithm. Before the detection module, it added the ViBe model to estimate the foreground target, which greatly reduces the detection area. The tracking module uses the Sift feature matching algorithm to replace the optical flow method in the original algorithm, accurately tracking the target to avoid the tracking drift, reducing the complexity of the calculation and improving the ability of the algorithm to adapt to the environment. The experiment results show that the improved TLD algorithm can improve the running speed, and the tracking accuracy can also be improved when the target is seriously occluded and the light intensity changes dramatically.