Monocular Long-Term Target Following on UAVs
Abstract
In this paper, we investigate the challenging long-term visual tracking problem and its implementation on Unmanned Aerial Vehicles (UAVs). By exploiting the inherent correlation between Frequency tracker And Spatial detector, we propose a novel tracking algorithm, denoted as FAST. As can be theoretically and analytically shown, the superior performance of FAST originates from: 1) robustness - by transforming from frequency tracker to spatial detector, FAST owns comprehensive detector to cover consequential temporal variance/invariance information that inherently retained in tracker, 2) efficiency - the coarse-tofine redetection scheme avoids the training of extra classifier and exhaustive search of location and scale. Experiments testified on tracking benchmarks demonstrate the impressive performance of FAST. In particular, we successfully implement FAST on quadrotor platform to tackle with indoor and outdoor practical scenarios, achieving real-time, automatic, smooth, and long-term target following on UAVs.
Cite
Text
Li et al. "Monocular Long-Term Target Following on UAVs." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2016. doi:10.1109/CVPRW.2016.11Markdown
[Li et al. "Monocular Long-Term Target Following on UAVs." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2016.](https://mlanthology.org/cvprw/2016/li2016cvprw-monocular/) doi:10.1109/CVPRW.2016.11BibTeX
@inproceedings{li2016cvprw-monocular,
title = {{Monocular Long-Term Target Following on UAVs}},
author = {Li, Rui and Pang, Minjian and Zhao, Cong and Zhou, Guyue and Fang, Lu},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
year = {2016},
pages = {29-37},
doi = {10.1109/CVPRW.2016.11},
url = {https://mlanthology.org/cvprw/2016/li2016cvprw-monocular/}
}