TY - JOUR
T1 - Performance evaluation of low resolution visual tracking for unmanned aerial vehicles
AU - Wang, Yong
AU - Wei, Xian
AU - Shen, Hao
AU - Hu, Jilin
AU - Luo, Lingkun
N1 - Funding Information:
This work was partially supported by National Science Found for Young Scholars under Grant No. 61806186, State Key Laboratory of Robotics and System (HIT) under Grant No. SKLRS-2019-KF-15, the program ‘Construction of Fujian Research Institute on Intelligent Logistics Industry Technology’ under Grant No. 2018H2001, CAS Pioneer Hundred Talents Program (Type C) under Grant No. 2017-122, and the program ‘Quanzhou Science and Technology Plan’ under Grant No. 2019C112, No. 2019C011R and No. 2019STS08.
Publisher Copyright:
© 2020, Springer-Verlag London Ltd., part of Springer Nature.
PY - 2021/4
Y1 - 2021/4
N2 - Several datasets for unmanned aerial vehicle (UAV) visual tracking research have been released in recent years. Despite their usefulness, whether they are sufficient for understanding the strengths and weakness of different resolution videos tracking remains questionable. Tracking in low resolution videos is a critical problem in UAV tracking. To address this issue, we construct a group of low resolution tracking datasets and study the performance of different trackers on these datasets. We find that some trackers suffered more performance degradation than others, which brings to light a previously unexplored aspect of the tracking methods. The relative rank of these trackers based on their tracking results on the datasets may change in the presence of low resolution. Based on these findings, we develop a multiple feature tracking framework which takes advantage of image enhancement scheme to improve image quality. In addition, we utilize the forward and backward tracking to evaluate multiple feature tracking results. Experimental results demonstrate that our tracker is competitive in performance to state-of-the-art methods in different resolutions scenarios. We believe our studies can provide a solid baseline when conducting experiments for low resolution UAV tracking research.
AB - Several datasets for unmanned aerial vehicle (UAV) visual tracking research have been released in recent years. Despite their usefulness, whether they are sufficient for understanding the strengths and weakness of different resolution videos tracking remains questionable. Tracking in low resolution videos is a critical problem in UAV tracking. To address this issue, we construct a group of low resolution tracking datasets and study the performance of different trackers on these datasets. We find that some trackers suffered more performance degradation than others, which brings to light a previously unexplored aspect of the tracking methods. The relative rank of these trackers based on their tracking results on the datasets may change in the presence of low resolution. Based on these findings, we develop a multiple feature tracking framework which takes advantage of image enhancement scheme to improve image quality. In addition, we utilize the forward and backward tracking to evaluate multiple feature tracking results. Experimental results demonstrate that our tracker is competitive in performance to state-of-the-art methods in different resolutions scenarios. We believe our studies can provide a solid baseline when conducting experiments for low resolution UAV tracking research.
KW - Fusion algorithm
KW - Image enhancement
KW - Low resolution dataset
KW - UAV tracking
UR - http://www.scopus.com/inward/record.url?scp=85087016717&partnerID=8YFLogxK
U2 - 10.1007/s00521-020-05067-3
DO - 10.1007/s00521-020-05067-3
M3 - Journal article
AN - SCOPUS:85087016717
SN - 0941-0643
VL - 33
SP - 2229
EP - 2248
JO - Neural Computing and Applications
JF - Neural Computing and Applications
IS - 7
ER -