A Benchmark and Simulator for UAV Tracking

Abstract

We present a benchmark and simulator for UAV tracking. Our benchmark comprises 70 video sequences captured from a UAV perspective and annotated with bounding boxes, covering diverse scenarios including bird-eye-view, moving camera, and varying altitude. We evaluate 16 state-of-the-art trackers on our benchmark and find that most existing algorithms perform significantly worse than on standard ground-level datasets, highlighting the unique challenges of aerial tracking. To support learning-based approaches, we develop a photo-realistic UAV simulator built on Unreal Engine that enables large-scale synthetic data generation with full annotation control.

Resources

Citation

@inproceedings{mueller2016benchmark,
  title     = {A Benchmark and Simulator for {UAV} Tracking},
  author    = {M{\"{u}}ller, Matthias and Smith, Neil and Ghanem, Bernard},
  booktitle = {European Conference on Computer Vision (ECCV)},
  year      = {2016}
}