Crowdsourcing Road Junction Maps With Three-dimensional Structure And Connectivity

Massive construction of road networks and constant changes in traffic conditions constitute a contemporary scenario of people’s daily travel,and trigger great demand for precise and up-to-date maps and information of roads,which poses new opportunities and challenges for road mapping technology.Junctions are the key nodes of urban road networks and characterized by intricate planar and vertical structures and flexible topological connectivity.With existing technology for professional road surveying,such as onboard mobile mapping systems,aerial photogrammetry,satellite-based remote sensing,it takes high costs to collect data and long periods to update maps,which is insufficient to meet the demand for data currency.Crowdsourced trajectory data are easy to collect and update at low costs and considerable coverage,and thus have been used to construct maps.At road junctions,this task remains challenging due to missing data,noisy elevation measurements,and changing topology.Following a framework of map data production consisting of data processing,information extraction,and dynamic updates,this study proposes approaches to junction mapping from crowdsourced trajectory data,which includes:(1)Due to the lack of models that are capable of trajectory reconstruction for long gaps,a probabilistic model based on historical patterns of space-time paths is developed to reconstruct missing trajectory episodes.In this model,historical travel paths and stops are encoded to extract movement patterns from trajectory archives.Stochastic diffusion in the directed random walk model is improved to bias towards these patterns,which also allows a long gap to be divided into a chain of multiple directed movement episodes rather than modeled as a single episode.For each episode,a time series of visit probability distribution can be estimated and sampled to reconstruct a trajectory gap.(2)In face of inter-series discrepancy and intra-series noise of the elevation measurements in crowdsourced trajectory data,a method for three-dimensional information extraction of complex junction structures is proposed based on the segmentation and fusion of trajectories.In this method,slope sections are located by finding trends in elevation series.Following the principle of consensus negotiation,the fusion of trajectory data under the restriction of continuous elevation variation produces a three-dimensional piecewise linear structure of a junction.This provides a solution to the problem of extracting three-dimensional geometry and elevation semantic information,and extends existing planar junction maps to a third dimension.(3)In order to mitigate the side effects of errors in traffic counts and uneven probe distribution on previous methods,we build an approach to change detection in junction connectivity by testing the count of turns dynamically.The probability distribution of traffic counts for a turn is estimated based on the assumption that all trips follow randomized shortest paths.A subsequent Poisson binomial test compares the actual traffic count for the turn against the estimation to identify anomalous values that indicate a closed or open turn status.This approach breaks limitations of threshold approaches including intolerance of count errors and the dichotomy of connectivity.