基于HOG-SVM的改进跟踪-学习-检测算法目标跟踪方法

对于经典TLD(跟踪-学习-检测)跟踪算法,在目标受到遮挡、光照、干扰、旋转和尺度变化等问题时,会导致算法的跟踪精度和速度降低,计算的复杂度较高,实时性差。针对以上问题,本文提出了一种改进的TLD目标跟踪算法。首先针对检测模块中计算复杂度高的问题,将HOG-SVM结合替换原TLD算法中的2bitBP特征和集成分类器；再针对原算法中跟踪精度低的问题,将KCF跟踪算法替换中值光流法；在HOG-SVM+KCF跟踪算法的基础上,对滑动窗口法进行改进,解决原算法中实时性差的问题。实验表明,改进后的跟踪算法,在背景环境变化的情况下,跟踪精度和速度都有提高,实时性加强。

Improved TLD Target Tracking Method Based on HOG-SVM

For the classical TLD (tracking-learning-detection) tracking algorithm, when the target is occluded, illuminated, disturbed, rotated and scaled, the tracking speed and accuracy of the algorithm will be reduced, the computational complexity is high, and the real-time performance is poor. To solve the above problems, an improved TLD target tracking algorithm is proposed in this paper. Firstly, aiming at the high computational complexity in the detection module, HOG-SVM is combined to replace the 2bitBP feature and integrated classifier in the original TLD algorithm; secondly, aiming at the low tracking accuracy in the original algorithm, KCF tracking algorithm is replaced by median optical flow method; on the basis of HOG-SVM+KCF tracking algorithm, sliding window method is improved to solve the problem of poor real-time performance in the original algorithm. Experiments show that the improved tracking algorithm improves the tracking accuracy and speed and enhances the real-time performance when the background environment changes.