Pavement macrotexture, quantified by mean texture depth (MTD) and mean profile depth (MPD), is a critical parameter for road safety and performance. The traditional sand patch test is labor-intensive and slow, creating a bottleneck for modern pavement management systems. Accurately translating the rich point cloud data into reliable MTD values using the 3D scanning method remains a challenge, with current methods often relying on oversimplified correlations. This research addresses this gap by developing and validating a novel machine learning framework to predict MTD and MPD directly from high-resolution 3D laser scans. A comprehensive dataset of 127 pavement samples was created, combining traditional sand patch measurements with detailed 3D point clouds. From these point clouds, 27 distinct surface features spanning statistical, spatial, spectral, and geometric domains were developed. Six machine learning algorithms, consisting of Random Forest, Gradient Boosting, Support Vector Regression, k-Nearest Neighbor, Artificial Neural Networks, and Linear Regression, were implemented. The results demonstrate that the ensemble-based Random Forest model achieved superior performance, predicting MTD with an R2 of 0.941 and a mean absolute error (MAE) of 0.067 mm, representing a 56% improvement in accuracy over traditional digital correlation methods. Model interpretation via SHAP analysis identified root mean square height (Sq) and surface skewness (Ssk) as the most influential features.
Richard et al. (Sun,) studied this question.