Keywords: train simulation framework, lidar, visual dataset, backscattered intensity, Lambertian–Beckmann model, navigation systems.

Guaranteeing safety and efficient traffic management in the railway network requires the correct execution of crucial and challenging tasks, such as localization, object detection, and obstacle avoidance. An increasing number of solutions are exploring the use of visual sensors to enhance accuracy without requiring a significant support at the infrastructure level. This paper proposes a simulation framework to generate LiDAR data for testing and validating novel algorithms in a railway scenario, where gathering these kinds of data in a real setting is impractical and time-consuming. Given a train trajectory, the framework exploits a graphic engine to generate the railway infrastructures in the surrounding area of the rail tracks, populating the virtual world with environmental objects. The point cloud of the LiDAR is generated through the ray-casting system of the graphic engine, taking into account the radiometric nature of the sensor, including the backscattered intensity for each point of the frame computed with the Lambertian–Beckmann model. Moreover, each LiDAR point comes with a semantic label that can be used to generate datasets for training and testing deep neural networks for object detection or segmentation tasks. The experiment results show the reliability of the LiDAR simulation to reproduce the sensor behaviour both in the distance measurements of the point cloud and in the backscattered intensity model.

download the full-text of this paper (PDF, 8 pages, 553 Kb)

go to the previous paper
go to the next paper
return to the table of contents
return to the volume description