Automated Tram Accessibility Assessment Using Digital Twin Simulations

About

This project develops a framework for automated assessment of tram accessibility for wheelchair users, based on the simulation of a Mobility Aid Lift (MAL) in a large-scale Digital Twin. Using a custom-built vehicle-mounted multi-sensor mapping system consisting of lidar, cameras, Global Navigation Satellite System (GNSS) and Inertial Navigation System (INS), the project team collected comprehensive geospatial data along selected tram routes in Melbourne. Digital twin assets were then produced as co-registered Cloud Optimised Point Clouds (COPC) and stereo-view imagery. We trained a deep learning model to detect tram tracks and tram stop flags from the stereo image sequences, projected to the 3D space of the digital twin. MAL deployment zones of pre-defined dimensions were then simulated relative to the tracks to assess the presence of physical obstacles preventing MAL deployment as well as ground gradient and camber. Benchmarked against visual inspections and manual measurements, the automated framework has shown an accuracy of around 90% in all of the assessment tasks, namely tram stop extraction, obstacle detection and ground gradient and camber estimation.