Determination of train positions within a railway network must be fail-safe and of high accuracy. In train-bourne positioning, exploitation of geometrical map features is an important factor and uncertainties in the map information may affect the position estimate. In this paper, we present a method to estimate the position of a train in the track net and to identify the correct path behind a turnout, using absolute position estimates and geometrical map information of various accuracies. We evaluate the impact of uncertainties in the map representation on the correct identification of a path behind a turnout. We derive a formulation of a probabilistic track map and include the map information into a constrained multi-hypothesis Kalman filter. We show in numerical simulations on a crossover and a turnout that modelling existing map uncertainties significantly improves the track discrimination.