A Map Matching Algorithm for Noisy, Low Frequent Public Transportation GPS Data

Abstract

Identifying the traveled road segments from raw GPS trajectories on a digital road network is known as the Map Matching. Map Matching becomes a challenging problem when the sparse geo-temporal data set is noisy (e.g., 10 meters away from the actual location) and has a low sampling rate (e.g., one data point per 3 minutes). The public transportation domain (e.g., buses) differs from the generic transportation (e.g., taxis) as it follows a predefined route, and that helps to build the ground truth trajectories. Ground truth trajectories are essential to validate the map-matching algorithms. There are many advanced map matching algorithms, but they are focused on the generic map matching problem. We propose an improvement to the existing Hidden Markov Model (HMM) map matching methodology to find the most likely road route considering the probability of the bus being on the predefined route. The proposed algorithm is validated using simulated GPS data in a dense road network with different noises and sample rates. Finally, the results are compared with the existing HMM solution using Route Mismatched Fraction (RMF).