LiDAR Based Landmark and Odometry Fusion for Precise Rail Localization Without Additional Infrastructure

This paper addresses the need for precise train localization within rail networks. Traditional methods face challenges such as high costs, imprecision, and vulnerabilities to disruptions. To overcome these issues, we propose a localization approach without additional infrastructure, i.e., using only existing infrastructure landmarks, specifically catenary poles, using a Deep Neural Network (DNN) within the 3D point cloud captured by a LiDAR sensor. By matching these landmarks to a precomputed two-dimensional HD Map and fusing this information with a state-of-the art LiDAR odometry algorithm, our proof of concept demonstrates the ability to achieve submeter precision for electrified rail segments.