摘要
在飞机等飞行目标的尺度以及视角发生变化时,核相关滤波(KCF)算法由于跟踪边框固定和滤波器准确性较低而易导致目标跟踪丢失。针对这一问题,在KCF算法基础上增加一种模型更新策略以提高模型准确性,并适时利用YOLOv5l检测网络实现对目标尺度的精确估计。在构建的飞机目标数据集上进行了实验验证,结果表明:相比原算法改进KCF算法在精确度和成功率上分别提升了0.315和0.285,在目标尺度及视角发生变化时具有较强的跟踪性能。
When the scale and viewing angle of flying targets such as aircraft change Kernelized Correlation Filtering(KCF)algorithm may cause target tracking loss due to fixed tracking boundary and low filtering accuracy.To solve this problem based on the KCF algorithm a model updating strategy is added to improve the accuracy of the model and the YOLOv5l detection network is used to achieve accurate estimation of the target scale.Finally the experimental results on the constructed aircraft target dataset show that the improved KCF algorithm has improved the accuracy and success rate by 0.315 and 0.285 respectively in comparison with the original algorithm and it has good tracking performance when the target scale and viewing angle change.
作者
杜鑫
沙建军
张祥
孙殿星
谭聪
DU Xin;SHA Jianjun;ZHANG Xiang;SUN Dianxing;TAN Cong(Qingdao Innovation and Development Base Harbin Engineering University,Qingdao 266000 China;No.59 Research Institute,China Ordnance Industry,Chongqing 401000 China)
出处
《电光与控制》
北大核心
2025年第3期27-32,共6页
Electronics Optics & Control
基金
航天发展基金。
关键词
飞行目标
目标跟踪
核相关滤波算法
尺度变化
视角变化
YOLOv5l
flying target
target tracking
kernelized correlation filtering algorithm
scale change
viewing angle change
YOLOv5l
