Answers are grounded in this paper and cite their sources.
Surface texture information is pivotal to driving safety and asphalt pavement condition evaluation. Conventional surveys rely on costly laser-based equipment, limiting large-scale application. This study proposes an efficient algorithm to reconstruct 3D asphalt pavement surfaces from multi-view images. It integrates SuperPoint with SuperGlue for feature extraction and matching, outperforming the traditional Scale-Invariant Feature Transform (SIFT) and Nearest-Neighbor Mutual (NN-Mutual) methods. A comprehensive validation framework compared algorithm performance, laser-scanned and Sand Patch Test measurements, and texture parameters. Results show SuperPoint detects more features on regular shapes, corners, and edges, while SIFT is more sensitive to pixel gradients. SuperGlue excels in complex textures matching, whereas NN-Mutual suits simpler point pairs. Image-based reconstructions achieved mean texture depth (MTD) within ±10% of laser measurements, with skewness and kurtosis differences within ± 0.1 and 0.3, respectively, demonstrating high fidelity. Consistent Kullback-Leibler divergence values using laser-scanned data as reference confirm that reconstructed surfaces preserve the distributional characteristics of real pavement. Overall, the proposed approach offers a cost-effective, accurate, and robust alternative for practical pavement texture evaluation, with strong potential for automated road inspection and intelligent construction applications.
No takes yet. Share an insight, caveat, or question.
Zhou et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: