Octree‐based point cloud simulation to assess the readiness of highway infrastructure for autonomous vehicles

Abstract

Autonomous vehicles (AVs) are anticipated to supersede human drivers with an expectation of improved safety and operation. Since current infrastructure is designed based on the constraints caused by human drivers, it must be reassessed for autonomous driving compatibility. Recently, representatives from the infrastructure owners/operators (IOOs), automotive industry, and academia have advocated for new approaches to prepare roadways for the deployment of AVs over the next decade. Following these recommendations, this paper proposes a novel, simulation‐based approach for the assessment of highways "readiness" for AVs using 3D point cloud data. The proposed method uses octrees to perform volumetric queries for potential obstructions within an AV sensory field. The proposed approach is compared to a state‐of‐the‐art raycasting approach. Consequently, available sight distances and maximum safe speed limits based on road and AV characteristics are proposed. Finally, a discussion of the potential mitigation measures at the locations with limited sight distances is presented.

Automated Object Detection, Mapping, and Assessment of Roadside Clear Zones Using Lidar Data

“…In Ural et al (30), a Digital Surface Model (DSM) was used in filtering buildings. DSMs, if used in a raycasting technique, can produce several false positives underneath elevated objects because of their form as a Triangular Irregular Network (TIN) (18,19). DSMs are also limited in their ability to represent overlapping objects when projected to a 2-D plane (18,19).…”

“…DSMs, if used in a raycasting technique, can produce several false positives underneath elevated objects because of their form as a Triangular Irregular Network (TIN) (18,19). DSMs are also limited in their ability to represent overlapping objects when projected to a 2-D plane (18,19). Therefore, DSMs are not adequate in accounting for vertical surfaces (elevation difference at the same horizontal location) and thus are considered a 2.5-D approach rather than 3-D (32).…”

“…All the vectors and the edge points were voxelated to improve performance when searching for the guard rails. Details about voxelization may be found in (18). Each voxel is a cube with a side-length of 0.12 m. After the normal vectors and the edge point's vectors were normalized, the voxels and the vectors created were used to search for obstacles.…”

“…To determine this, transportation agencies have traditionally needed to manually document roadside clear zone size and slope, and the position of obstructions on a case-bycase basis. Recently, however, because of its millimeterlevel accuracy, reduction in cost, and widespread data collection capabilities, civil engineering research has shifted to the use of lidar as a fast and feasible alternative for data collection (14)(15)(16)(17)(18)(19)(20).…”

“…Lidar technology uses high-frequency laser scanners to map its surroundings by creating high-resolution 3-D point clouds (19,21,22). Additional data, such as scan angle and intensity of the returned light beam, are also recorded (18). A specific variant of this technology known as Mobile Laser Scanning (MLS) can be used for efficient recording of road and roadside data (23).…”