Developing a Method to Automatically Extract Road Boundary and Linear Road Markings from a Mobile Mapping System Point Cloud Using Oriented Bounding Box Collision-Detection Techniques

Abstract

Advancements in data-acquisition technology have led to the increasing demand for high-precision road data for autonomous driving. Specifically, road boundaries and linear road markings, like edge and lane markings, provide fundamental guidance for various applications. Unfortunately, their extraction usually requires labor-intensive manual work, and the automatic extraction, which can be applied universally for diverse curved road types, presents a challenge. Given this context, this study proposes a method to automatically extract road boundaries and linear road markings by applying an oriented bounding box (OBB) collision-detection algorithm. The OBBs are generated from a reference line using the point cloud data’s position and intensity values. By applying the OBB collision-detection algorithm, road boundaries and linear road markings can be extracted efficiently and accurately in straight and curved roads by adjusting search length and width to detect OBB collision. This study assesses horizontal position accuracy using automatically extracted and manually digitized data to verify this method. The resulting RMSE for extracted road boundaries is +4.8 cm and +5.3 cm for linear road markings, indicating that high-accuracy road boundary and road marking extraction was possible. Therefore, our results demonstrate that the automatic extraction adjusting OBB detection parameters and integrating the OBB collision-detection algorithm enables efficient and precise extraction of road boundaries and linear road markings in various curving types of roads. Finally, this enhances its practicality and simplifies the implementation of the extraction process.