Benchmarking Image Perturbations for Testing Automated Driving Assistance Systems

TL;DR

This work addresses the vulnerability of vision-based ADAS perception to image perturbations by benchmarking 32 perturbations drawn from 8 categories, across two representative ADAS tasks (semantic segmentation and LK/ACC) and two simulators. It introduces PerturbationDrive, a modular library enabling offline and online robustness testing, and demonstrates that perturbations reveal robustness failures at both component and system levels. The study shows that perturbation-based data augmentation and continuous learning substantially improve generalization to unseen weather domain perturbations, though they may increase driving jitter, highlighting a trade-off between robustness and control smoothness. Overall, the paper provides a practical framework for systematic ADAS robustness evaluation and domain-adaptation strategies with clear implications for real-world deployment and future research directions.

Abstract

Advanced Driver Assistance Systems (ADAS) based on deep neural networks (DNNs) are widely used in autonomous vehicles for critical perception tasks such as object detection, semantic segmentation, and lane recognition. However, these systems are highly sensitive to input variations, such as noise and changes in lighting, which can compromise their effectiveness and potentially lead to safety-critical failures. This study offers a comprehensive empirical evaluation of image perturbations, techniques commonly used to assess the robustness of DNNs, to validate and improve the robustness and generalization of ADAS perception systems. We first conducted a systematic review of the literature, identifying 38 categories of perturbations. Next, we evaluated their effectiveness in revealing failures in two different ADAS, both at the component and at the system level. Finally, we explored the use of perturbation-based data augmentation and continuous learning strategies to improve ADAS adaptation to new operational design domains. Our results demonstrate that all categories of image perturbations successfully expose robustness issues in ADAS and that the use of dataset augmentation and continuous learning significantly improves ADAS performance in novel, unseen environments.

Figures (2)

• Figure 1: Valid and invalid perturbation types.
• Figure 2: Benchmarking perturbations.