Due to the proposed method's unique features, the lanes' width 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.
(Papers: ANT-2020, Trans. Res. Part C-2021)
A sample of the reconstructed RDM for a test network around Antwerp, Belgium, can be seen by clicking this link.
The construction of routable digital maps based on trajectory data has attracted much 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. This research presents a three-step automatic method based on QuickBundles for node detection in the road network, a dissimilarity matrix based on Fréchet 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 proposed method's accuracy, connectivity, compatibility, validity, and robustness have been tested in various ways. The results of this research show that this method, while 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.