Road Intersection Extraction Based on Low-Frequency Vehicle Trajectory Data

Abstract

Global navigation satellite system (GNSS) vehicle trajectory data play an important role in obtaining timely urban road information. However, most models cannot effectively extract road information from low-frequency trajectory data. In this study, we aimed to accurately extract urban road network intersections and central locations from low-frequency GNSS trajectory data, and we developed a method for accurate road intersection identification based on filtered trajectory sequences and multiple clustering algorithms. Our approach was founded on the following principles. (1) We put in place a rigorous filtering rule to account for the offset characteristics of low-frequency trajectory data. (2) To overcome the low density and weak connection features of vehicle turning points, we adopted the CDC clustering algorithm. (3) By combining the projection features of orientation values in 2D coordinates, a mean solving method based on the DBSCAN algorithm was devised to obtain intersection center coordinates with greater accuracy. Our method could effectively identify urban road intersections and determine the center position and more effectively apply low-frequency trajectory data. Compared with remote sensing images, the intersection identification accuracy was 96.4%, the recall rate was 89.6%, and the F-value was 92.88% for our method; the intersection center position’s root mean square error (RMSE) was 10.39 m, which was 14.9% higher than that of the mean value method.