针对Staple算法存在的两个问题, 提出了一种基于Staple改进的目标跟踪算法。首先, 为了增强Staple算法在灰度视频序列中的判别能力, 提出一种基于局部敏感直方图的直方图分类器; 其次, 提出一种基于相对置信度的自适应融合系数, 解决了Staple算法中两个分类器无法最优融合的问题。将该算法在OTB2013测试集上与其他9个先进的算法进行了比较, 实验结果表明该算法的精确度与准确率分别为0.814和0.614, 相对于Staple算法分别提升了4.1%和3.5%, 具有很好的鲁棒性。

Focusing on the two problems existed in Staple tracker, an improved target tracking algorithm based on Staple tracker is proposed. Firstly, in order to strengthen the discriminative ability of Staple tracker for gray video sequences, a histogram classifier based on locality sensitive histogram is proposed. Furthermore, as the constant merge factor in Staple tracker can not merge two classifiers optimally, an adaptive merge factor is proposed by using the relative confident coefficient. Experiment is made on OTB2013 benchmark for comparing the proposed tracker with other 9 state-of-the-art trackers. The results show that the proposed tracker has a precision of 0.814 and a success rate of 0.614, improved by 4.1% and 3.5% respectively compared with that of the Staple tracker, which indicates that the proposed tracker is more robust.