Search for a command to run...
The reliable operation of camera-based advanced driver-assistance systems in nighttime road traffic presents a significant challenge. Driving in low light impairs visibility, increases noise, and introduces complex interactions with glare and motion. Traditionally, automotive lighting design optimises headlamps for the human visual system, which acts as a passive observer. However, camera sensors differ significantly in spectral sensitivity, dynamic range, and noise behaviour. As a result, lighting and sensing systems are often optimised in isolation, neglecting their fundamental photometric interaction. This gap motivates a systematic investigation that treats the headlamp as an active, optimisable component of the machine perception stack. The objective of this thesis is to quantify the photonic relationship between vehicle headlamp light distributions and the signal quality achieved by forward-facing cameras. Methods are established to evaluate and optimise headlamp systems from a perception-oriented perspective, focusing on how specific beam patterns influence the signal-to-noise ratio (SNR) and contrast reproduction in safety-critical Regions of Interest (RoI) under nighttime conditions. A structured field test analyses the influence of different low-beam light distributions on camera signal quality. The setup includes multiple cameras, calibrated reflectance targets positioned at defined distances, and humanoid dummies. SNR is derived by mapping measured scene luminances through laboratory-calibrated photon-transfer curves, while contrast reproduction is assessed using Weber contrast and Kolmogorov–Smirnov tests. Complementarily, a simulation-based benchmark integrates a validated camera model with a safety-derived RoI based on speed and time-to-collision (TTC). Deep learning models are evaluated under controlled noise to derive spatially varying SNR requirements. The benchmark metric, coverage, quantifies the fraction of the RoI that meets this SNR. The results demonstrate that the spatial light distribution is the dominant factor determining camera SNR. The best-performing light distribution achieves 2.6 times higher coverage than the worst, highlighting the strong impact of illumination geometry on signal quality. In a complementary parameter study, the influence of sensor settings is analysed through systematic sweeps of key camera parameters. The results establish a clear hierarchy of influence: Pixel Size > Exposure Time ≈ Aperture F-number > Read Noise, confirming that variations in sensor configuration cannot compensate for an unfavourable light distribution. Coverage correlates significantly (r = 0.70) with the human-centric Headlamp Safety Performance Rating (HSPR). Contrast analyses show that maximum headlamp intensity is not universally optimal, and in the presence of ambient lighting, adaptive dimming improves contrast similarity. Beyond static illumination effects, dynamic factors in nighttime traffic introduce additional degradation mechanisms related to motion and glare. Motion leads to spatial smearing of image content, primarily affecting the lateral periphery of the RoI, while glare induces flare propagation and a reduction in local contrast. The analysis quantifies these effects and proposes mitigation strategies that combine short-exposure HDR branches with spatially adaptive illumination. These findings underline the importance of adaptive light distributions that dynamically balance exposure, contrast, and glare suppression to maintain stable perception performance under varying driving conditions. This thesis demonstrates that spatial light distribution exerts a decisive and quantifiable influence on camera-based perception at night. The findings establish the headlamp as a functional element of the perception stack and motivate a co-design paradigm in which lighting and sensing are optimised jointly. The developed methods provide a transferable foundation for perception-linked lighting evaluation, sensor-aware illumination control, and the development of robust, adaptive headlamp systems for automated driving.