Multiple model estimation scheme for map-matching

A map-matching method using a multiple model (MM) estimation scheme is presented in this paper. A vehicle travelling on the road network is subject to often changes in the movement direction as the vehicle turns onto another road at an intersection or negotiates a road bend. Hence, the estimation of the successive locations of the vehicle on the road network from a sequence of noisy position measurements can be accomplished by a MM filter with a time-varying set of switching models. Each model describes a rectilinear movement along a road which is likely to be used by the vehicle at the current instant. As underlying MM estimation technique, the proposed method applies a variable structure variant of the latest change moment testing algorithm (VS-LCMT) whereby hypotheses about the moment of the last sudden change of the road model are successively generated. The current road is identified by finding the road model which has associated the most probable hypotheses and the estimate of the vehicle state is obtained by probabilistically combining the partial estimates provided by a bank of filters matched to these hypotheses. The imminence of a road model change is quantified on the basis of the state-dependent probability of the vehicle being outside the current road. This imparts flexibility to the algorithm. The algorithm allows the reduction of the computational effort by utilizing 1D state estimates.