An investigation of detecting potholes with UAV LiDAR and UAV Photogrammetry

Sammanfattning: Potholes are caused by erosion and as such always emerging on our roadnetwork. Potholes may not only cause great damages to vehicles, but can alsocause road accidents, which in the worst case are fatal. Today, the detection ofpotholes is usually based on citizen reports or ocular inspection by vehicle,where a loose description of the potholes properties and location can be given.Recent research has explored the possibility of aerial inspection of paved roadswith the new, cost effective, Structure-from-Motion (SfM) technique, whichcan produce 3D point clouds from photogrammetric data. SfM point cloudshave then been used in conjunction with processing algorithms toautomatically detect and extract potholes from paved surfaces. However, theresults have not been optimal for practical use. The purpose of this study is,therefore, to explore the possibility of using UAV LiDAR for potholedetection in paved roads as a better alternative to the currently popularStructure-from-Motion (SfM) technique. A LiDAR point cloud is derived by alaser scanner and may have several advantages over SfM, for instance, theinsensitivity to poor light conditions and modelling errors. This study is setout to answer how point clouds derived from UAV SfM and UAV LiDARcompare to each other regarding detecting potholes of different sizes, wheredetected potholes will be compared to ground truth data. An elevation check,consisting of 126 height control points along the paved road, will also be usedto evaluate the height accuracy in the clouds. Data collection is done with theUAV system mdLiDAR3000DL aaS containing a RIEGL miniVUX-1DLlaser scanner for LiDAR data and Sony RX1R II 42.4 megapixel camera forSfM data. The data for both methods are collected during the same flight. Theproposed method automatically detects and extracts potholes from a pavedsurface based on the vertical distance to local reference planes which representthe undamaged road surface. The point clouds are filtered in CloudComparebefore imported to TerraScan for detection and extraction of potholes. Theextraction results are then controlled by a set of terrestrial measurements bytotal station. The results show that potholes with a smaller width of at least16.5 cm and a depth of at least 2.7 cm can be detected and extracted frompoint clouds derived by UAV LiDAR at a flight altitude of 30 m. Theextracted potholes had a standard deviation of 1.40 cm in width and 6.7 mmin depth. Shadows on the road caused height anomalies in the point cloudproduced by Structure-from-Motion (SfM), which made pothole detectionimpossible with the proposed methodology.