Extraction of road surface from mobile LiDAR data of complex road environment

Three-dimensional (3D) data of roadways are frequently used for extraction of detailed roadway information which is essential for several planning and engineering applications. Recent past has seen rapid growth in utilization of mobile LiDAR system (MLS) to acquire volumetric 3D data of roadway for this purpose. MLS data are capable of capturing highly detailed road information, which is useful for road maintenance and road safety operations. The existing literature shows that road environment complexity, unevenness, and absence of raised curb limit the extraction of road information from MLS data. It must be noted that a large number of roads, especially in developing world, are characterized by these complexities and thus raise the need for a technique which can work in these road environments. Considering the above, this paper proposes a method to extract road information, where road boundary is not geometrically well-defined. The proposed method is constructed using unstructured MLS data as input and does not require any other additional data. The method is divided into three major steps, that is, MLS data structuring and ground filtering, road surface point extraction, and road boundary refinement. The first step filters ground points from input MLS data, while the second step identifies road surface points from among the ground points. The second step is designed using specific characteristics of a road, that is, topology, surface roughness, and variation of point density. Third step refines road boundary. Three test sites, quite complex with heterogeneous characteristics, were used for demonstration of the proposed method. Road surfaces of these three roadways were accurately extracted without being affected by on-road objects and absence of raised curb. Average accuracy measures like completeness, correctness, and quality were found to be 93.8%, 98.3%, and 92.3%, respectively, in three test sites. Further, road boundaries of extracted road surfaces of these three test sites were refined at average completeness, correctness, and quality of 95.6%, 97.9%, and 93.7%, respectively. The proposed method has shown satisfactory performance for complex roadways having road section with and without raised curb, and has potential to be employed for such road environments, which are not uncommon. Proposed method was implemented on GPU-based parallel computing framework, which significantly saved the run time in processing of MLS data of three test sites.