Evaluating map matching algorithms for smartphone GNSS data: matching vehicle trajectories to an urban road network

Data from vehicles tracked using Global Navigation Satellite Systems (GNSS) can be used to monitor driving behaviour and road safety. In usage-based insurance programs, driver insurance premiums are tailored according to individual driving behaviour, often using data collected from user-owned smartphones. Due to positional noise caused, map matching algorithms must be used to spatially link GNSS observations to the road network. The purpose of this study is to evaluate the performance of several algorithms to process smartphone GNSS data for vehicular trips in urban road networks. This study evaluated five implementations, namely one topological (TMM), two probabilistic (PMM) using Hidden Markov Models (HMM), one fuzzy (FMM), and one hybrid map matching algorithm (HyMM) in terms of match accuracy and run time. Data was collected using ten smartphone devices and three applications across 12 trip scenarios in Montreal, Canada targeting the downtown, old city, highways, bridges, and tunnels. Results were compared with a series of ANOVA tests. Accuracy was not significantly different for the best performing algorithms (the Fast HMM and TMM) followed by the HyMM, with the Standard HMM and FMM algorithms performing significantly worse. Only the FMM algorithm was significantly slower than the others in terms of run time.