An Integrated LFM/LDC/RSU Positioning Method for Autonomous Vehicles

This paper proposes a vehicle positioning method that combines premade lidar feature maps (LFM), lateral distance constraints (LDC), and a single roadside unit (RSU). Firstly, we perform registration between the current lidar frame and the premade lidar feature maps to calculate the point-to-line and point-to-plane residuals. Secondly, we utilize a monocular camera to detect the adjacent lane line and obtain the lateral distance observation between the vehicle and the adjacent lane line with a similar relationship. A lateral distance residual is calculated by comparing the visual lateral distance observation, which significantly reduces the positioning error. Thirdly, we utilize a single RSU to observe the distance between the RSU and the vehicle. A further single RSU distance residual is calculated by comparing the RSU distance measurement, effectively reducing the positioning error. Then, we figure out the total residual based on the above residuals and solve the optimization equation with Ceres to obtain the vehicle position. Finally, experimental results show that the RMSE is less than 10 cm in the campus scene and demonstrate that the proposed method can improve vehicle positioning in sparse lidar feature regions.