基于TLD和fDSST的长时间目标跟踪算法

A Long-Term Tracking Algorithm Based on TLD and fDSST

为加强fDSST算法在目标快速运动、快速形变、目标消失情况下的跟踪精度,提出了一种基于TLD和fDSST的长时间目标跟踪算法。在fDSST算法的基础上,加入了检测器和学习器对跟踪结果进行修正和学习,并利用检测器和学习器的正负样本对跟踪结果进行置信度评估,从而解决了在跟踪失败情况下的错误参数学习问题。实验表明,基于TLD和fDSST的长时间目标跟踪算法,不但解决了fDSST算法由于目标快速运动、形变甚至消失而使跟踪失败,难以进行长时间持续跟踪的问题,且很大程度上增强了TLD算法的跟踪精度。

In order to improve the tracking precision of the f DSST algorithm when the target is moving fast deforms or even disappears a long-term tracking algorithm based on TLD and f DSST is proposed. Based on the f DSST algorithm a detector and an online learner are used to modify and learn from the tracking results. To solve the problem of learning wrong parameters when the f DSST algorithm fails the positive and negative samples of the detector and the learner are used to evaluate the tracking results. The experiment results indicate that the long-term tracking algorithm based on TLD and f DSST solves the problem that the f DSST algorithm fails to track the target for a long time when the target is in fast moving deforms or even disappears. This approach could also improve the tracking precision of the TLD algorithm.