RMR-Net: Degradation-Evidence-Guided Road-Image Restoration for Defect Detection

2026-08-09

AI Summary

This study addresses the challenge of degraded in-vehicle road images—commonly affected by motion blur, defocus, low illumination, and noise—which obscure the boundaries of pavement defects such as cracks and potholes and thereby impair detection performance. To tackle this, the authors propose RMR-Net, a lightweight task-aware restoration network that innovatively integrates a degradation evidence estimation mechanism and a bounded detail residual pathway. Without requiring real-world ground-truth degradation data, RMR-Net leverages synthetically generated degradation parameters to guide conditional restoration, effectively recovering high-frequency details. Experiments demonstrate that RMR-Net achieves state-of-the-art mAP50 on seven out of eight degradation types across the IVCNZ and PCM datasets, notably improving to 0.427 under motion blur on IVCNZ (baseline: 0.140) and 0.233 under defocus on PCM (baseline: 0.060), substantially enhancing the performance of the downstream YOLOv11s detector.

Abstract

Vehicle-mounted road cameras are vulnerable to motion blur, defocus, poor illumination, and noise, which can erase thin cracks and pothole boundaries needed by road defect detectors. This paper presents RMR-Net, a compact task-aware restoration front end that estimates degradation evidence from the image, optionally fuses it with existing corruption context/parameters, conditions lightweight restoration blocks, and returns high-frequency pavement detail through a bounded residual path. The experimental scope is deliberately controlled: the conditioning information used on the Image and Vision Computing New Zealand (IVCNZ) pothole dataset and the Road Damage Dataset: Potholes, Cracks and Manholes (PCM) consists of saved synthetic-generator parameters, not measured vehicle telemetry. A clean-trained, frozen YOLO11s detector evaluates every image source. Across eight held-out degradation conditions, RMR-Net obtains the highest mAP50 in seven, including 0.140-0.427 for IVCNZ motion blur and 0.060-0.233 for PCM defocus. A compact ablation identifies the bounded detail path as the largest local contributor, while degradation conditioning and detector-aware stability terms provide complementary guidance.

Problem

• road-image restoration
• degradation
• defect detection
• motion blur
• defocus

Innovation

• degradation-evidence-guided restoration
• bounded residual path
• task-aware image restoration
• road defect detection
• lightweight restoration network