Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Common Corruptions for Enhancing and Evaluating Robustness in Air-to-Air Visual Object Detection

Anastasios Arsenos, Vasileios Karampinis, Evangelos Petrongonas, Christos Skliros, Dimitrios Kollias, Stefanos Kollias, Athanasios Voulodimos

TL;DR

The paper introduces AOT-C, the first robustness benchmark for air-to-air visual object detection, by applying seven common corruptions to the AOT dataset to simulate challenging flight conditions. It conducts a large-scale evaluation of eight monocular detectors—spanning one-stage YOLO variants and multi-stage transformers—revealing that one-stage detectors exhibit stronger robustness to corruption while transformer-based methods are more vulnerable. The authors implement a synthetic corruption augmentation strategy and show that finetuning on corrupted data improves generalization to real-world flight scenarios, addressing domain shift. These findings guide detector selection and augmentation strategies for robust see-and-avoid in autonomous aerial systems.

Abstract

The main barrier to achieving fully autonomous flights lies in autonomous aircraft navigation. Managing non-cooperative traffic presents the most important challenge in this problem. The most efficient strategy for handling non-cooperative traffic is based on monocular video processing through deep learning models. This study contributes to the vision-based deep learning aircraft detection and tracking literature by investigating the impact of data corruption arising from environmental and hardware conditions on the effectiveness of these methods. More specifically, we designed $7$ types of common corruptions for camera inputs taking into account real-world flight conditions. By applying these corruptions to the Airborne Object Tracking (AOT) dataset we constructed the first robustness benchmark dataset named AOT-C for air-to-air aerial object detection. The corruptions included in this dataset cover a wide range of challenging conditions such as adverse weather and sensor noise. The second main contribution of this letter is to present an extensive experimental evaluation involving $8$ diverse object detectors to explore the degradation in the performance under escalating levels of corruptions (domain shifts). Based on the evaluation results, the key observations that emerge are the following: 1) One-stage detectors of the YOLO family demonstrate better robustness, 2) Transformer-based and multi-stage detectors like Faster R-CNN are extremely vulnerable to corruptions, 3) Robustness against corruptions is related to the generalization ability of models. The third main contribution is to present that finetuning on our augmented synthetic data results in improvements in the generalisation ability of the object detector in real-world flight experiments.

Common Corruptions for Enhancing and Evaluating Robustness in Air-to-Air Visual Object Detection

TL;DR

The paper introduces AOT-C, the first robustness benchmark for air-to-air visual object detection, by applying seven common corruptions to the AOT dataset to simulate challenging flight conditions. It conducts a large-scale evaluation of eight monocular detectors—spanning one-stage YOLO variants and multi-stage transformers—revealing that one-stage detectors exhibit stronger robustness to corruption while transformer-based methods are more vulnerable. The authors implement a synthetic corruption augmentation strategy and show that finetuning on corrupted data improves generalization to real-world flight scenarios, addressing domain shift. These findings guide detector selection and augmentation strategies for robust see-and-avoid in autonomous aerial systems.

Abstract

The main barrier to achieving fully autonomous flights lies in autonomous aircraft navigation. Managing non-cooperative traffic presents the most important challenge in this problem. The most efficient strategy for handling non-cooperative traffic is based on monocular video processing through deep learning models. This study contributes to the vision-based deep learning aircraft detection and tracking literature by investigating the impact of data corruption arising from environmental and hardware conditions on the effectiveness of these methods. More specifically, we designed types of common corruptions for camera inputs taking into account real-world flight conditions. By applying these corruptions to the Airborne Object Tracking (AOT) dataset we constructed the first robustness benchmark dataset named AOT-C for air-to-air aerial object detection. The corruptions included in this dataset cover a wide range of challenging conditions such as adverse weather and sensor noise. The second main contribution of this letter is to present an extensive experimental evaluation involving diverse object detectors to explore the degradation in the performance under escalating levels of corruptions (domain shifts). Based on the evaluation results, the key observations that emerge are the following: 1) One-stage detectors of the YOLO family demonstrate better robustness, 2) Transformer-based and multi-stage detectors like Faster R-CNN are extremely vulnerable to corruptions, 3) Robustness against corruptions is related to the generalization ability of models. The third main contribution is to present that finetuning on our augmented synthetic data results in improvements in the generalisation ability of the object detector in real-world flight experiments.
Paper Structure (18 sections, 1 equation, 2 figures, 3 tables)

This paper contains 18 sections, 1 equation, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Aerial Object Detection
  4. Robustness Benchmarks
  5. The Proposed Synthetic Dataset
  6. Corruption Setup
  7. Discussion on the disparity between synthetic and real-world corruptions
  8. Experimental Setup
  9. Benchmark Dataset
  10. Real-world Flight tests
  11. Metric setup
  12. Evaluation Results
  13. Robustness Evaluation on AOT Dataset
  14. Corruptions Types that Affect Aerial Detection
  15. Network Attributes Affecting Robustness
  16. ...and 3 more sections

Figures (2)

  • Figure 1: Visualization of the seven corruption types for each severity level in our benchmark
  • Figure 2: Qualitative evaluation of YOLOv5 object detector on real flights when using corruptions as augmentation (Finetuned model) or not (Base model)