Real-Time Map Matching: Enhancing Urban Fleet Localization on Edge Devices Using Standard GPS

Sammanfattning

Accurately tracking commercial vehicle fleets is crucial but challenging in dense urban environments due to so-called multipath errors in Global Positioning System (GPS) signals. Current systems often rely on offline processing, which prevents the real-time insights required for safety monitoring and eco-driving. This thesis presents a robust real-time map matching algorithm based on a Hidden Markov Model (HMM), specifically designed to run locally on resource-constrained edge devices. By relying exclusively on GPS telemetry and the Swedish National Road Database (NVDB), the model integrated directional and speed penalties alongside a sliding window Viterbi decoder to mitigate signal disturbances without the use of internal vehicle sensors, such as Inertial Measurement Units (IMU). The algorithm was evaluated under simulated hardware constraints for the Eurotech DynaGATE 10- 14, against a historical dataset of 170 public transit trips from which severe anomalies and incomplete routes had been systematically filtered out. The results indicate that accurate real-time vehicle localization on edge devices is feasible in the evaluated setting and shows comparable accuracy to the offline baseline.