Description

The emergence of high-resolution automotive grade lidar sensor technologies and the development of advanced deep- learning frameworks open the door for accurate and real-time perception of the surroundings of mobile trains for improved safety and operational efficiency. Under this project, we developed key components and proof-of-concepts of lidar-based multimodal solutions that rely on advanced deep- learning methods and utilize object detection approaches to demonstrate the feasibility of real-time detection of objects and events that could represent major safety issues for moving trains. The detected objects could be vehicles or people trespassing a railroad track, or they could be hazardous objects such as falling rocks obstructing a moving train. The project developed multiple deep -learning– based software algorithms that demonstrate the feasibility of a robust real-time railroad safety system. The project also integrated Next-Generation automotive-grade lidars and cameras, which are readily deployed by Autonomous Vehicles (AVs) and Advanced Driver Assistance Systems (ADAS), with deep learning-based object detection solutions to demonstrate the feasibility of achieving high levels of object detection accuracy and robustness under challenging weather conditions. Using automotive-grade sensor technologies can lead to significant savings in cost for the railroad industry while enabling the deployment of accurate real-time detection solutions that have been improving over the past decade, and which continue to improve as new sensors, deep- learning architectures, and other AI solutions emerge.

Publication Date

9-11-2026

Publication Type

Report

Topic

Miscellaneous, Transit and Passenger Rail, Transportation Technology

Digital Object Identifier

10.31979/mti.2026.2412

MTI Project

2412

Keywords

Railroad safety, Lidar-based object detection, Multimodal perception, Multimodal fusion, Cooperative perception, Generative models, Track obstruction

Disciplines

Transportation

Download

Share

COinS