摘要
本文提出一种融合的长时单目标跟踪策略,融合基于深度学习的目标检测算法和基于相关滤波的单目标跟踪算法,形成两种跟踪模式,根据多种判断策略控制跟踪模式切换,能够在保证跟踪鲁棒性的情况下大幅度提高跟踪实时性。将该跟踪策略用在自制数据集ATP-UAV上,取得了较好的跟踪效果。
This paper proposes a fused long-term single object tracking strategy that combines deep learning-based object detection algorithms with correlation filter-based single object tracking algorithms.This fusion creates two tracking modes that can be switched based on multiple decision-making strategies.It significantly improves tracking real-time performance while ensuring robustness.The proposed tracking strategy is applied to the self-made ATP-UAV dataset and achieves excellent tracking results.
作者
陈毅夫
王琳琳
施嘲风
孟凡军
杨霄
任伟杰
CHEN Yi-fu;WANG Lin-lin;SHI Chao-feng(AVIC Beijing Precision Engineering Institute for Aircraft Industry,Beijing 100076;Beihang University,Beijing 100043)
出处
《航空精密制造技术》
2023年第4期45-47,62,共4页
Aviation Precision Manufacturing Technology
关键词
深度学习
相关滤波
长时单目标跟踪
融合策略
deep learning
correlation filtering
long-term single object tracking
fusion strategy