Beyond Mode Detection: Reconstructing Detailed Transit Itineraries from Crowdsourced GPS Trajectories

Reconstructing detailed public transit itineraries from GPS trajectories is crucial for urban transportation planning, yet remains challenging in dense transit networks due to spatiotemporal complexities. Current approaches primarily focus on broad transportation mode detection (walk/bus/rail), failing to identify specific transit lines or stop-level behaviors, which is a critical limitation for operational analytics and equity assessment. This paper presents two novel algorithms to address this gap: a GTFS-based method employing an R-tree spatial index to precisely match trajectory segments with transit routes, and a Google Maps Directions API-based approach leveraging iterative leg matching for route inference. Both algorithms share a unified preprocessing pipeline for trajectory cleaning and staypoint detection, enabling identification of high-resolution transit patterns. When evaluated on crowdsourced GPS data from Baltimore City’s transit system, our GTFS-based algorithm achieved ~80% accuracy in identifying specific transit lines, while the API-based method reached ~85%, significantly outperforming conventional mode detection approaches. These methods also extract new metrics such as actual transfer wait times, boarding/alighting patterns, and route-specific travel durations, which enable targeted service optimizations and deeper transit equity analyses beyond theoretical models. By converting raw GPS data into actionable insights, these algorithms provide urban planners with critical tools to optimize transit service in complex multi-modal networks.