Academic papers documenting theoretical frameworks, experimental approaches, and comprehensive research findings.
This thesis addresses the critical challenge of pothole detection under adverse weather conditions where traditional detectors experience up to 90.6% accuracy degradation. We propose an enhancement-aware domain-generalized approach using modern YOLO detectors (YOLO11, YOLOv26) with condition-aware enhancement modules, achieving 401% accuracy recovery on night conditions while maintaining real-time performance suitable for ITS applications.
Bučko, B., et al. (2022) - Computer Vision Based Pothole Detection under Challenging Conditions. Sensors, 22(22), 8878. DOI: 10.3390/s22228878
This foundational study demonstrates severe performance degradation of pothole detectors under challenging conditions. Night conditions show catastrophic 90.6% accuracy drop, rain conditions 34.5% degradation, and sunset conditions 31.4% degradation. Serves as the baseline for our enhancement-aware approach, highlighting the critical need for condition-aware preprocessing in road damage detection systems.
Clear: 1,052 images | Rain: 286 images | Sunset: 184 images | Evening: 196 images | Night: 118 images