基于遮挡判断的自适应更新相关滤波跟踪算法  

An Adaptive Updating Correlation Filter Tracking Algorithm Based on Occlusion Judgment

作　　者：郭勇[1,2] 赖广 GUO Yong;LAI Guang(State Key Laboratory of High Performance Complex Manufacturing,Central South University,Changsha 410000,China;Sunward Intelligent Equipment Co.Ltd.,Changsha 410000,China)

机构地区：[1]中南大学高性能复杂制造国家重点实验室,长沙410000 [2]山河智能装备股份有限公司,长沙410000

出　　处：《电光与控制》2021年第12期57-60,66,共5页Electronics Optics & Control

摘　　要：针对传统的核相关滤波跟踪算法缺乏处理目标存在遮挡情况的能力,提出了遮挡判断指标以及模型自适应更新的改进算法。首先通过最大响应值和低响应点个数两个指标综合判断是否存在遮挡,然后自适应调整模型学习率,解决了存在遮挡时不能准确跟踪的问题。在OTB2015数据集中选取存在遮挡的图像序列验证了算法的性能,相比传统的核相关滤波算法对遮挡情况下的跟踪,精确度提高了15.12%,成功率提高了14.7%。实验结果表明改进后的算法在存在遮挡时能够准确跟踪目标,具有更高的准确率和鲁棒性。The traditional Kernel Correlation Filtering(KCF)tracking algorithm lacks the ability to deal with target occlusion.To solve the problemocclusion judgment indexes and an improved algorithm with adaptive model updating are proposed.Two indexesmaximum response value and the number of low response pointsare used to comprehensively determine whether there is occlusionand then the model's learning rate is adjusted adaptivelyso as to avoid inaccurate tracking in the presence of occlusion.The image sequences with occlusion are selected from OTB2015 data set to verify the performance of the algorithm.Compared with that of the traditional KCF algorithm for tracking under occlusionthe accuracy is increased by 15.12%and the success rate is increased by 14.7%.The experimental results show that the improved algorithm can accurately track targets when there is occlusion with higher accuracy and stronger robustness.

关 键 词：目标跟踪 相关滤波 遮挡判断 阈值 学习率 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]