Navigation in constrained areas such as ports or dense urban environments is often exposed to global navigationsatellite system (GNSS) satellites masking caused by the infrastructures. In this case, the GNSS positioning is inaccurate orunavailable and proprioceptive sensors are generally used to temporarily localise the vehicle on a map. However, the drift ofthese sensors rapidly causes the navigation system to fail. In this study, the navigation is computed using magnetometer andGNSS observations defined in an absolute reference frame. The heading measurements are coupled with a digital map of theroad network in order to localise the vehicle in a map matching process. The contribution of this study is the proposition of aparticle filter that fuses the position and direction observations to estimate the vehicle position. In this context, when the GNSSsignal is masked, the observations of direction are used to compute the vehicle position. The proposed filter is defined in bothcircular and linear domains in order to take into account the nature of the observations. The proposed approach is assessed onreal and synthetic data.