| 三联神经网络与区域自适应策略融合的目标跟踪方法 |
| |
| 引用本文： | 王建中,张驰逸,孙庸.三联神经网络与区域自适应策略融合的目标跟踪方法[J].北京理工大学学报,2021,41(2):169-176. |
| |
| 作者姓名： | 王建中 张驰逸 孙庸 |
| |
| 作者单位： | 北京理工大学机电学院,北京 100081 |
| |
| 基金项目： | 国家部委基础科研计划资助项目（JCKY2019602C015） |
| |
| 摘 要： | 为解决目标跟踪过程中快速运动模糊、背景相似干扰、目标状态变化等问题，基于孪生网络跟踪算法，提出三联区域候选神经网络（TripleRPN）算法与跟踪区域自适应策略（TAA）相融合的目标跟踪方法（TAA+TripleRPN）.三联区域候选神经网络根据当前跟踪结果实时更新网络匹配模板，提高了跟踪器对目标状态变化的敏感性.通过区域自适应策略，根据区域候选回归网络分类分支的得分在网络的两组输出间择优选择，提高算法长时跟踪的鲁棒性.针对背景相似干扰和目标状态变化的问题时，TAA+TripleRPN跟踪器能达到更好的跟踪性能.在OTB2015数据集上，算法的AUC达到66.31%，CLE达到88.28%.在实际场景中实现验证与应用，跟踪效果良好.
|
| 关 键 词： | 目标跟踪 深度学习 三联区域候选回归神经网络 |
| 收稿时间： | 2020/6/8 0:00:00 |
Target Tracking Method Based on Fusion of Triple Neural Network and Area Adaptation |
| |
| WANG Jianzhong,ZHANG Chiyi,SUN Yong.Target Tracking Method Based on Fusion of Triple Neural Network and Area Adaptation[J].Journal of Beijing Institute of Technology(Natural Science Edition),2021,41(2):169-176. |
| |
| Authors: | WANG Jianzhong ZHANG Chiyi SUN Yong |
| |
| Affiliation: | School of Mechatronical Engineering, Beijing Institute of Technology, Beijing 100081, China |
| |
| Abstract: | In order to solve the problems of fast motion blur, background similar interference and target state change in the process of target tracking, a target tracking method (TAA+TripleRPN) that combines the triple area candidate neural network (tripleRPN) algorithm with the tracking area adaptive strategy (TAA) was proposed based on siamese network tracking algorithm. The triple-area candidate neural network updates the network matching template in real time based on the current tracking results, which improves the sensitivity of the tracker to changes in the target state. Through the regional adaptive strategy, based on the scores of the classification candidates of the regional candidate regression network, the two groups of network outputs are selected optimally, which improves the robustness of the algorithm''s long-term tracking. For the problems of similar background interferences and target state changes, the TAA+TripleRPN tracker can achieve better tracking performance. On the OTB2015 dataset, the algorithm has an AUC of 66.31% and a CLE of 88.28%. The verification and application are implemented in actual scenarios, and the tracking effect is good. |
| |
| Keywords: | object tracking deep learning TripleRPN |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《北京理工大学学报》浏览原始摘要信息 |
|
点击此处可从《北京理工大学学报》下载全文 |
|
