Lane-level routable digital map reconstruction for motorway networks using low-precision GPS data

The construction of routable digital maps based on trajectory data has attracted a lot of attention, especially in recent years, with the ease and cheapness of collecting the required data. Such maps, if they are constructed at lane-level, have many applications in traffic analysis, especially the study of driving behavior based on floating car data. In this paper, we present a three-step automatic method based on QuickBundles for the node detection in the road network, a dissimilarity matrix based on Fre ' chet distance for road centerline construction, and the Gaussian Mixture Method for lane estimation. The results are a smooth, segment-based centerline unbiased by GPS density distribution over lanes with accurate road width as well as compatible and highly accurate estimation of lanes. The accuracy, connectivity, compatibility, validity, and robustness of the proposed method have been tested in various ways. The results of this paper show that this method, while being low cost, can construct accurate lane-level routable digital maps that can be used as a platform for extracting longitudinal and lateral driving behavior, especially drivers' lane-changing maneuvers. Due to the unique features of the proposed method, the width of the lanes remains constant along the entire length of the road segments. In addition, the validation of the method based on traffic metrics shows that the constructed maps can be used to obtain reliable estimates of the speed and volume of traffic flow in different lanes. Being derived from actual trajectories, the inferred lane markings may deviate from the physical ones if the driver population systematically deviates laterally, for instance in curves; depending on the application use case of the resulting map, this may be a desirable or undesirable feature.