Automated assessment of road marking condition is essential for data-driven pavement maintenance, yet existing methods either rely on large annotated datasets or lack interpretable degradation grading. This paper proposes a training-free framework that quantifies road marking deterioration through three physics-informed indicators: the Luminance Loss Index (LLI), which captures intensity attenuation; the Crack Index (CI), which measures surface fracturing; and the Spalling Area Ratio (SAR), which reflects structural loss relative to the original marking geometry. The three indicators are fused into a composite degradation score S via weighted linear combination, and a grid search procedure optimises the fusion weights and classification thresholds to assign each marking segment to one of three condition grades (Normal, Minor Damaged, or Damaged) without requiring annotated training data. Evaluated on a dataset of 1,672 line segment instances under 5-fold cross-validation, the proposed method achieves an accuracy of 0.7111 ± 0.0265 and a macro-F1 of 0.7175 ± 0.0247, outperforming an Otsu thresholding baseline by 7.7 percentage points on both metrics. Ablation experiments confirm that SAR provides the strongest discriminative signal, while CI improves Damaged class recall by 6.6 percentage points over the best two-indicator combination (LLI+SAR), reducing the risk of overlooking severely degraded markings. The framework requires no deep learning infrastructure and produces interpretable condition grades suitable for operational road safety management in scenarios where large annotated datasets are unavailable.
Xiaotian Jiang (Thu,) studied this question.