Reliable Perception under Veiling Glare: Synthetic Glare Generation and Realism Validation for Automated Vehicles

Veiling glare caused by sunlight critically compromises camera-based perception in Automated Vehicles (AVs) by reducing image contrast and visibility. This phenomenon, particularly prominent during daytime driving, degrades object and lane detection performance and undermines safety-critical AV decision-making. Although several studies have addressed glare detection and mitigation, the limited availability and diversity of real-world glare datasets hinder the development and validation of automotive perception systems under such challenging conditions. This paper introduces a synthetic veiling glare injection approach for camera-based AV datasets, enabling controlled synthesis of glare effects by parameterizing glare shape, size, position, color, number, intensity, and transparency to realistically emulate sun-induced veiling glare. Two multi-metric evaluations are presented to assess the perceptual, photometric, and physical consistency of the injected glare through comparative analyses: (i) synthetic glare versus glare-free data, and (ii) synthetic glare versus naturally captured glare from the same dataset. The comparative results demonstrate that the proposed injection process effectively reproduces the dominant characteristics of realworld veiling glare with mild to moderate severity, showing strong agreement in metric behavior. Consequently, the proposed injection and evaluation framework provides a validated, glare-aware foundation for safety assessment in automated driving systems.

Keywords: synthetic glare, veiling glare, glare injection, SSIM, PSNR, Contrast difference, spatial frequency power difference.