多特征融合的抗干扰目标跟踪算法研究
Research on Anti-Jamming Target Tracking Algorithm Based on Multi-Feature Fusion
黄朝锋 1葛海波 1周婷 1李强1
作者信息
- 1. 西安邮电大学电子工程学院,陕西 西安 710121
- 折叠
摘要
针对目标跟踪在遮挡和背景复杂等干扰导致的准确率较低的问题,提出了一种多特征融合的目标跟踪算法.算法采用加权系数的方法将HOG特征、CN特征和ULBP特征自适应的融合各个特征分量信息,对检测到的遮挡目标进行加权滤波重定位,以自适应模型更新策略解决跟踪目标在遮挡时导致的跟踪模板被污染的问题.在公开数据集OTB100 上进行了实验验证,与传统KCF跟踪算法对比,上述算法在精确度和正确率分别提升了 16.2%和 14.9%,证明了上述算法可以更加准确的完成跟踪任务.
Abstract
Aiming at the problem of low accuracy of target tracking caused by interference such as occlusion and complex background,this paper proposes a multi-feature fusion target tracking algorithm.The algorithm adopts the method of weighting coefficient to adaptively fuse the information of each feature component of the HOG feature,CN feature and ULBP feature,performs weighted filtering and relocation on the detected occluded target,and uses an a-daptive model update strategy to solve the problem of tracking target caused by occlusion.The tracking template is polluted.The experimental verification was carried out on the public data set OTB100,and compared with the tradi-tional related filtering algorithms MOSSE algorithm,CN algorithm,CSK algorithm,KCF algorithm and deep learning algorithm C-COT algorithm and ECO algorithm respectively.The simulation results show that the algorithm in this pa-per is:compared with the MOSSE algorithm,the accuracy and accuracy are increased by 39%and 24.5%,compared with the CN algorithm by 43.9%and 26.8%,compared with the CSK algorithm by 26.4%and 21.4%,compared with the KCF algorithm.Compared with the C-COT algorithm,it is 1.7%and 3%higher than the C-COT algorithm.Although it is slightly worse than the ECO algorithm,the algorithm in this paper has more advantages in processing speed.
关键词
目标跟踪/多特征融合/加权滤波/模型更新/深度学习Key words
Target tracking/Multi-feature fusion/Weighted filtering/Model update/Deep learning引用本文复制引用
基金项目
陕西省自然科学基金(2011JM8038)
陕西省重点产业创新链(群)项目(S2019-YF-ZDCXL-ZDLGY-0098)
西安邮电大学研究生创新基金(CXJJLY202050)
西安邮电大学研究生创新基金(CXJJYL2021007)
出版年
2024
NETL
NSTL
万方数据