[BA/IDP] Real-Time Assessment of Camera Image Quality for Autonomous Vehicles Under Varying Weather and Lighting Conditions

Institut

Lehrstuhl für Fahrzeugtechnik (TUM-ED)

Typ

Bachelorarbeit /

Inhalt

 

Beschreibung

BACKGROUND

Camera-based perception systems are a core component of modern autonomous vehicles. Object detection, lane keeping, and scene understanding often rely on high-resolution camera images, whose quality, however, is strongly affected by external influences. Rain, fog, snow, glare, dusk, or sudden lighting changes (e.g., entering or exiting a tunnel) can significantly degrade image quality.

To ensure the safety of autonomous driving functions, it is necessary to continuously monitor camera image quality in real time. Reliable image quality assessment allows the system to react early to quality degradation.

Classical image quality metrics (e.g., BRISQUE, NIQE, PIQE) as well as newer learning-based approaches offer different trade-offs in terms of accuracy, computational efficiency, and robustness against automotive-specific disturbances. Since no reference image is typically available in the automotive context, no-reference approaches (NR-IQA) are of particular interest. However, a systematic investigation into which metrics are suitable for use under real, dynamic weather and lighting conditions is still missing.

Voraussetzungen

YOUR ROLE

• Literature review of existing image quality metrics (classical and learning-based), with a focus on no-reference methods and their application in the automotive context
• Analysis and selection of suitable datasets with varying weather and lighting conditions (e.g., DAWN, ACDC, BDD100K, nuScenes, or custom data collection)
• Implementation and/or adaptation of selected image quality metrics
• Development of an evaluation pipeline to systematically assess the metrics with regard to:
  • Sensitivity to weather and lighting influences
  • Correlation with the performance of downstream perception tasks (e.g., object detection)
  • Computational efficiency and real-time capability on embedded/edge hardware
• Comparative evaluation of the investigated metrics and derivation of recommendations
• Optional: development or adaptation of a custom metric or a lightweight model for real-time use
• Documentation of results in the written thesis and presentation of findings

WHAT YOU SHOULD BRING ALONG

• Strong Motivation and Interest for AVs
• Basic Knowledge in Programming, e. g. Python, C++
• Structured and independent way of working

If you are interested in joining this project, feel free to send me an application with your CV and transcript of records. I look forward to receive your application.

Tags

FTM Studienarbeit, FTM AV, FTM AV Safe Operation, FTM Karunainayagam, FTM Krau

Möglicher Beginn

sofort

Kontakt

Nijinshan Karunainayagam, M. Eng.
Raum: MW3507
Tel.: +49 89 289 15386
nijinshan.karunainayagamtum.de