Automatic classification of urban pavements using mobile LiDAR data and roughness descriptors

Maintenance of urban road pavements is shown as critical in city management. Mobile Laser Scanning is well-established for achieving geo-reference point clouds and imagery depicting the reality in an accurate and productive manner and intense efforts have been made to implement automatic methods for extracting geometric and semantic useful information. This work shows a method based on evaluating roughness descriptors from mobile LiDAR data to automatically segment and classifies asphalt and stone pavements. The method includes segmentation of 1 m slices from the road, extraction of profiles, evaluation of roughness parameters, and K-means clustering. Among the evaluated roughness descriptors, only arithmetic average of absolute values (between 2.5 · 10−3 m to 4 · 10−3 m for stone and 10−3 to 2 · 10−3 m for asphalt) and root mean squared (between 3 · 10−3 m to 5 · 10−3 m for stone and 10−3 m to 2 · 10−3 m for asphalt) show robust classification. The methodology is tested in three real case studies.