3D Pavement Surface Reconstruction Based on K-Nearest Neighbor Filtering Algorithm

Based on the principles of three-dimensional laser scanning, this study conducted comprehensive acquisition, denoising, dataset generation, and 3D reconstruction of surface data for asphalt pavement specimens of various types. First, rutting slabs made of different asphalt mixtures were prepared, and surface 3D data were collected using a Leica laser scanning system. Due to contamination and stains inherent to the asphalt specimens, the raw scanned data contained noise. To address this, a K-nearest neighbor (KNN)-based filtering algorithm grounded in machine learning was employed for denoising. An adaptive parameter selection strategy was integrated into the algorithm design to mitigate the impact of subjective human factors on filtering performance. To meet the input data structure requirements of asphalt pavement texture depth estimation algorithms based on different principles, the 3D surface data were transformed into depth matrices. Additionally, data augmentation techniques were applied to expand the point cloud dataset. As a result, two types of 3D pavement datasets-depth matrix-based and point cloud-based-were constructed, enabling high-fidelity 3D reconstruction of pavement surfaces. This work provides essential data support for the development of diverse asphalt pavement texture depth measurement algorithms.