Robust objectness tracking with weighted multiple instance learning algorithm
Robust objectness tracking with weighted multiple instance learning algorithm
复制标题
使用加权多实例学习算法进行鲁棒的对象跟踪
DOI：
10.1016/j.neucom.2017.02.106
复制
发表时间：
2018-05-02
期刊：
影响因子：
6
通讯作者：
Zheng, Zunxin
中科院分区：
文献类型：
--
作者：
Yang, Honghong;Qu, Shiru;Zheng, Zunxin
提出了一种用于视觉跟踪的新型改进在线加权多示例学习算法（IWMIL）。在IWMIL算法中，基于具有对象属性（跨超像素）的对象性估计来评估每个样本对包概率的贡献重要性。为了降低计算成本，采用了一种由粗到精的样本检测方法来检测新到达帧的样本。然后，提出了一种自适应学习率，它利用最大分类器得分来为跟踪结果和模板分配不同的权重，以更新分类器。此外，使用一种对象相似性约束策略来估计跟踪漂移。在具有挑战性的序列上的实验结果表明，所提出的方法对遮挡和外观变化具有鲁棒性。© 2018爱思唯尔有限公司。保留所有权利。
A novel improved online weighted multiple instance learning algorithm(IWMIL) for visual tracking is proposed. In the IWMIL algorithm, the importance of each sample contributing to bag probability is evaluated based on the objectness estimation with object properties (superpixel straddling). To reduce the computation cost, a coarse-to-fine sample detection method is employed to detect sample for a new arriving frame. Then, an adaptive learning rate, which exploits the maximum classifier score to assign different weights to tracking result and template, is presented to update the classifiers. Furthermore, an object similarity constraint strategy is used to estimate tracking drift. Experimental results on challenging sequences show that the proposed method is robust to occlusion and appearance changes. (C) 2018 Elsevier B.V. All rights reserved.