Problem
Real-time pothole detection must work under varied weather, lighting, and road-surface conditions, where visual noise can reduce the reliability of standard object-detection models.
Back to research
Results in Engineering · Volume 28 · December 2025
Robust Multi-Weather Pothole Detection
An open-access research paper by Shahnaj Parvin, Foysal Munsy, Md Tanzeem Rahat, Aminun Nahar, Kamruddin Nur, and Debasish Ghose on enhanced YOLOv9 for road-safety computer vision.
Why It Matters
Reliable pothole detection supports road safety, infrastructure monitoring, maintenance planning, and scalable intelligent transportation systems.
Core Idea
The paper improves YOLOv9 for pothole detection by modifying architecture components and training on the Multi-Weather Pothole Detection dataset with robustness-oriented augmentation.
Main Result
The improved YOLOv9 achieved an average mAP@50 of 95% and an F1-score of 91% on MWPD, outperforming the baseline YOLOv9 model.
What Was Difficult
The hard part is making object detection robust across multi- weather visual variation while keeping the model efficient enough for practical road-safety deployment.
Resources
Open Directions
- How well do pothole detectors generalize across cities, countries, and road textures?
- Can weather-aware augmentation reduce the need for large labeled road datasets?
- What model size is needed for reliable mobile or edge deployment?