Extracting road turns and intersections from crowd source GPS tracks

Abstract

We propose a novel method for extracting road intersections from user contributed GPS tracks. Our method is different from related work in that it deals with multiple issues related to real datasets, specifically noise, inconsistent and rather low sampling rate, and the difficulty of tuning parameters. We extract both intersections and turns, allowing applications to make better use of such GPS data. Firstly individual tracks are simplified, yielding turns. Then the samples are allowed to move closer to the real positions of the turns to cluster more accurately. Finally a progressive clustering is applied to detect turns and intersections. The experiments are done on a real dataset to evaluate both the convenience of the three processing steps, and the overall utility of our method in comparison to an often cited method.