Road extraction from low-cost GNSS-device dense trajectories

Abstract

This paper proposes a method for road centerline extraction from dense Global Navigation Satellite System (GNSS) trajectories, collected by using low-cost GNSS-devices, i.e. smartphones. The proposed method basically consists in generating a frequency image by tracking the GNSS trajectories and then by applying the Steger line detector to the generated image to extract the road centerlines. The main motivation of using the Steger algorithm is its capability to detect lines with sub-pixel accuracy. To evaluate the obtained results, reference road centerlines are manually extracted from a georeferenced orthomosaic. The experiments performed demonstrate the high potential of applying the Steger line detector to frequency images, generated by using dense GPS (Global Positioning System) trajectories. The completeness and correctness values for the accomplished experiments were 98% and 99%, respectively. Additionally, the RMSE (Root Mean Square Error) ranged from 0.63 m to 2.39 m, or approximately 1/16 to 1/4 of the expected accuracy (about 10 m) of a point determined by the Single-Point Positioning (SPP) method, which is the GNSS positioning method usually employed by smartphones.