Research On Special Road Network Structure Information Extraction Based On GPS Trajectory

High-Definition(HD)maps are in the early stages of development,which put forward higher requirements for road maps in terms of accuracy,completeness and richness of details.Road Information Extraction is one of the key process in map production.At present,the road objects of recognition and extraction include the following: the coordinates and structures of intersections,geometry of lane lines and lane centerlines in road segments,attribute information such as lane number,lane width,and lane boundaries in road segments,traffic signs,toll booth and other road furniture.While few researchers concentrate on special detailed road structures in road segments,as well as fusing extracted road information with existing maps to improve map geometric and topological information.GPS trajectory is an important dataset of road information extraction while their disadvantages are obviously.The uncertainty in quantity and accuracy of GPS trajectory bring certain challenges to road information extraction research.Additionally,the density of these datasets is spatially unevenly distributed.To address the issues mentioned above,the main work of this paper follows:(1)A method to detect and extract U-TURNs in road segments is proposed.In a road segment,a U-turn directly affects the topological association between the lanes,which in turn affects the way the road segment is stored in the map database.According to the heading of GPS points before and after passing U-TURNs,the turning point pairs in the road segment are extracted,and the projection mapping the heading angle to a2 D plane is realized to get the value of turning angle.The geometric position and turning heading of U-TURNs can be extracted by clustering the turning points.The experiment show that this method can effectively identify U-turns in road segments,and it also has good performance in three-dimensional urban roads.(2)Propose a method of identifying the main and auxiliary roads,as well as detecting the entrance and exit of auxiliary roads.Based on the distribution of density of trajectory points on road sections,the kernel density estimation method is used to obtain the trajectory density distribution curve.Then we extract relevant features,establish a classification model with Support Vector Machine(SVM).An optimization process is utilized in terms of the continuity of the road.Based on the results of continuous sections in road segments,different entrance and exit extraction strategies are adopted for different types of auxiliary roads.Experiments show that this method can effectively identify the auxiliary roads and the entrance and exit of auxiliary roads with a high recognition accuracy.(3)Propose a method to improve and enhance the existing networks to include the detailed roadway structures extracted above.In view of the situation that the research of road information extraction and the construction of complete road network is relatively fragmented,we adopt Open Lane Model to form the map database,update and improve geometric,topological and semantic information of main and auxiliary roads and U-turns by matching and fusion.