DrIFT: Autonomous Drone Dataset with Integrated Real and Synthetic Data, Flexible Views, and Transformed Domains

TL;DR

The paper introduces DrIFT, a comprehensive drone detection dataset with diverse domain shifts and a novel uncertainty metric, improving unsupervised domain adaptation for visual drone detection.

Contribution

It provides a new dataset with multiple domain shifts, background segmentation maps, and a novel uncertainty metric, enhancing domain adaptation methods.

Findings

DrIFT enables better evaluation of drone detection under domain shifts.

MCDO-map outperforms traditional uncertainty estimation methods.

Unsupervised domain adaptation with MCDO-map achieves state-of-the-art results.

Abstract

Dependable visual drone detection is crucial for the secure integration of drones into the airspace. However, drone detection accuracy is significantly affected by domain shifts due to environmental changes, varied points of view, and background shifts. To address these challenges, we present the DrIFT dataset, specifically developed for visual drone detection under domain shifts. DrIFT includes fourteen distinct domains, each characterized by shifts in point of view, synthetic-to-real data, season, and adverse weather. DrIFT uniquely emphasizes background shift by providing background segmentation maps to enable background-wise metrics and evaluation. Our new uncertainty estimation metric, MCDO-map, features lower postprocessing complexity, surpassing traditional methods. We use the MCDO-map in our uncertainty-aware unsupervised domain adaptation method, demonstrating superior performance to SOTA unsupervised domain adaptation techniques. The dataset is available at: https://github.com/CARG-uOttawa/DrIFT.git.