Lane-Level Network Information Extraction Method Based On Crowdsourced Trajectory Data

Urban road network data needs to be closely integrated with the fast-developing urban construction process,and accurate lane level road network information is crucial for the refined management of intelligent transportation system and residents’ travel navigation.Traditional lane level road network information extraction methods are based on traditional mapping and remote sensing images,which have high data collection costs and long update cycles,making it difficult to meet the demand for finegrained traffic control under the rapid development of cities.Crowdsourced trajectory data based on user sharing provides a new solution for fine-grained and fast information extraction of large-scale road networks by crowdsourcing the trajectory information of traffic participants.This thesis combined advanced technologies such as machine learning,deep learning,and data mining to conduct research on large-scale,fine-grained lane level road network information extraction.The main research contents are as follows:(1)Addressing the problems of uncertainty of crowdsourced trajectory data positioning,uneven distribution area,and data heterogeneity,this thesis studies the outlier detection method based on crowdsourced trajectory data,and proposes a spatialtemporal cascade detection framework.This framework detects location offset points in the spatial distribution based on the adaptive density clustering method,and detects motion outliers based on the SE-TCN-AE model to extract the input features in the temporal dimension.(2)Studying the identification of urban road planar intersections and their spatial structure extraction method according to the function of road intersections and the turning characteristics reflected by the trajectory data.This involves extracting the turning points in the trajectory data by analyzing the turning behavior,identifying the central location and coverage of each intersection using the aggregation pattern of the turning points,and extracting the spatial structure inside each intersection by obtaining the exit and entrance points and the driving direction attributes of vehicles driving into and out of the intersection range.(3)Investigating the method of extracting lane level road information based on low-cost and high presentational crowdsourced trajectory data,in view of the existing problems of high cost and long cycle time in obtaining lane level road information using remote sensing images and Li DAR data.This involves clustering trajectory data on the same road section based on the similarity of distance and direction between subtrajectory segments,to extract the road centerline and road boundary,and using the Gaussian distribution mixture model to detect the number of lanes,lane centerline,and lane width based on the spatial structure distribution characteristics of the trajectory data in the road plane and cross-section.