VISTA-RM: Vision-Based Integrated Strategy for Threshold Assessment of Road Markings in Autonomous Driving

Road-marking visibility is crucial for autonomous driving systems, yet current evaluation methods mainly rely on manual inspection of the retroreflective coefficient, R L , which is not well aligned with the perceptual characteristics of machine vision and fails to meet the requirements for real-time and accurate assessment. To address the gap in rapid and accurate visibility assessment, this study proposes a machine vision–based method for evaluating road-marking visibility. A contrast ratio (CR) metric and a brightness critical deviation (BCD) model are introduced to quantify road-marking visibility under different luminance and retroreflective conditions. Through perspective transformation, brightness features of lane markings were extracted to compute CR values. Subsequently, a BCD curve was constructed to evaluate lane-marking visibility. For markings with CR values below the threshold, a Contrast-Limited Adaptive Histogram Equalization (CLAHE) strategy was applied to improve local contrast before detection. Field-test results showed that the proposed method accurately identified failed markings, with an overall evaluation accuracy exceeding 0.9 across different detection models and illuminance levels. Compared with the traditional R L -based inspection method, the BCD-based method provided stronger consistency with machine-vision detection results, yielding an eightfold reduction in processing time and a 270.3% increase in data coverage. The results demonstrated that the proposed method enabled rapid, continuous, and perception-oriented assessment of road-marking visibility for autonomous driving applications.