Avoiding explicit map-matching in vehicle location

Abstract

In this paper we combine Kalman filters and Markov models to solve the problem of locating a vehicle travelling on a road network. The location system incorporates information from a noisy vehicle positioning system, a simple model of vehicle dynamics and driver behaviour, and a representation of the road network. The Markov model is used to handle the topological aspects of the problem, maintaining a set of hypotheses for the segment on which the vehicle is travelling and their respective probabilities. The Kalman filters handle the metric aspects, providing estimates of the vehicle location on each of the hypothesized road segments. The two are closely coupled, with the statistics from the Kalman filters used to update the Markov belief state at each time step, and the Markov model providing a probability distribution over the Kalman filters. This differs from conventional vehicle location schemes by not performing any explicit map-matching step, and has advantages in both robustness and flexibility.