IMAGE QUALITY CHALLENGES IN AUTONOMOUS DRIVING: A CONCEPTUAL FRAMEWORK FOR CAMERA LENS DEFECT DETECTION

Axmedov A. G. & Dadaxanov M. X.


Annotatsiya

Camera-based perception is fundamental for autonomous driving, yet its reliability is compromised by lens impairments. Temporary soiling (raindrops, mud, dust) and permanent defects (scratches, cracks) can obscure critical regions of the field of view, degrade detection accuracy, and create safety hazards. Although existing research addresses soiling detection, limited attention has been given to permanent lens damage and, more critically, to system-level safety responses when impairments occur. This paper proposes a safety-aware conceptual framework for lens defect detection and response. The framework integrates: (i) boot-time self-checks with automated cleaning, (ii) real-time monitoring that characterizes severity, criticality, and recoverability of impairments, (iii) decision policies that map impairments to graded responses (from alerts to cautious driving or full stop), and (iv) optional temporal inpainting for non-critical regions. Safety monitoring ensures that fallback sensors (radar, ultrasonic, LiDAR) are prioritized when visual confidence is reduced. The framework is dataset-agnostic, aligns with ISO 26262 and ISO 21448 standards, and is supported by example modules, state diagrams, and evaluation metrics. By linking lens defect detection with safety engineering, it provides both a theoretical foundation and a pathway toward practical deployment in autonomous vehicles.


Adabiyotlar

  1. [1] Uřičář, M., Vobecky, A., Sistu, G., & Yogamani, S. (2019). SoilingNet: Soiled pixel detection on automotive surround-view cameras. Electronic Imaging, *2019*(11), 298–1. https://doi.org/10.2352/ISSN.2470-1173.2019.11.IPAS-298
  2. Das, A., Uřičář, M., & Franke, U. (2020). TiledSoilingNet: Efficient real-time detection of lens soiling on automotive cameras. IEEE Transactions on Intelligent Vehicles, *5*(4), 614–623. https://doi.org/10.1109/TIV.2020.2992447
  3. Soboleva, E., & Shipitko, O. (2021). Raindrops on windshield: Dataset and lightweight gradient-based detection algorithm. IEEE Access, *9*, 160515–160526. https://doi.org/10.1109/ACCESS.2021.3132012
  4. WoodScape Consortium. (2019). WoodScape: A multi-task, multi-camera fisheye dataset for autonomous driving [Dataset]. https://woodscape.valeo.com
  5. Uřičář, M., Sistu, G., Rashed, H., Vobecky, A., Křížek, P., & Yogamani, S. (2019, October). Restoring soiled areas on automotive fisheye cameras. In *2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)* (pp. 2552–2560). IEEE. https://doi.org/10.1109/ICCVW.2019.00313
  6. International Organization for Standardization. (2018). Road vehicles — Functional safety (ISO Standard No. 26262).
  7. International Organization for Standardization. (2022). Road vehicles — Safety of the intended functionality (ISO Standard No. 21448).
  8. Kurup, A., & Petridis, V. (2024). Impact of raindrops on camera-based object detection in software-defined vehicles. In *Proceedings of the 2024 ACM/IEEE Symposium on Edge Computing* (pp. 112–125). IEEE.
  9. Axmedov, A., & Dadaxanov, M. (2024). ADVANCEMENTS IN IMAGE QUALITY ASSESSMENT: A COMPREHENSIVE SURVEY. DIGITAL TRANSFORMATION AND ARTIFICIAL INTELLIGENCE, 2(5), 34–39. https://dtai.tsue.uz/index.php/dtai/article/view/v2i56
  10. Axmedov Abdulazizxon Ganijon O‘g‘li, & Dadaxanov Musoxon Xoshimxonovich. (2025). ENHANCING AUTONOMOUS DRIVING SYSTEMS WITH AI-BASED IMAGE QUALITY ASSESSMENT FOR LENS DEFECT DETECTION. https://doi.org/10.5281/zenodo.15569355