Speaker
Description
The accurate and cost-effective acquisition of rut surface morphology data is regarded as crucial for rapid and high-frequency inspections for the road network-level. However, RGB-D sensors utilizing infrared structured light are affected by natural light interference, resulting in low precision despite cost advantages. Moreover, evaluating sampling points based on road cross-sections makes it challenging to precisely locate rut positions, and a single rut depth index fails to fully capture the overall rut morphology. This study proposes a method for the characterization and localization of rutting morphology by three-dimensional (3D) road surface reconstruction using RGB-D sensor. The approach begins by aligning RGB images with depth maps to generate 3D color point clouds. Depth information loss is reduced using an improved adaptive joint bilateral filtering method, which ensures edge-preserving denoising and hole repair. A point cloud stitching method combining sample consensus initial alignment (SAC-IA) and trimmed iterative closest point (ICP) is applied to asphalt pavements to overcome the depth camera's limited field of view and achieve comprehensive 3D reconstruction. Data from non-deformable areas are further used to construct the horizontal axis of the road surface, enabling the extraction of wheel rut endpoints and valley bottom points from refined point clouds for accurate measurement and lateral positioning. The experimental results demonstrate high correlation coefficients of 0.95 and 0.98 between the lateral positioning achieved by the proposed method and the manually measured rutting and valley bottom points, respectively. This method enables rapid and high-frequency inspection of asphalt pavement rutting damage for the road network-level.
