Probabilistic Map-Matching for Low-Frequency GPS Trajectories

The ability to infer routes taken by vehicles from sparse and noisy GPS data is of crucial importance in many traffic applications. The task, known as map-matching, can be accurately approached by a popular technique known as ST-Matching. The algorithm is computationally efficient and has been shown to outperform more traditional map-matching approaches, especially on low-frequency GPS data. The major drawback of the algorithm is a lack of confidence scores associated with its outputs, which are particularly useful when GPS data quality is low. In this paper, we propose a probabilistic adaptation of ST-Matching that equips it with the ability to express map-matching certainty using probabilities. The adaptation, called probabilistic ST-Matching (PST-Matching) is inspired by similarities between ST-Matching and probabilistic approaches to map-matching based on a Hidden Markov Model. We validate the proposed algorithm on GPS trajectories of varied quality and show that it is similar to ST-Matching in terms of accuracy and computational efficiency, yet with the added benefit of having a measure of confidence associated with its outputs.