Urban and Rural Road Extraction From Smartphone-Based GPS Trajectories

This paper presents an automatic method for extracting roads from dense set of GPS trajectories obtained through low-cost smartphone devices. In order to perform the experiments, GPS trajectories were converted into frequency images through a Kernel Density Estimator, followed by the application of an image thresholding method for eliminating low frequency pixels and then by using a skeletonization method for extracting the road axes. The experimental evaluation of obtained results takes into account the influence of the GPS trajectory density and the frequency image resolution, on the accuracy of the extracted road axes. The results were evaluated qualitatively by visual inspection and quantitatively by analyzing the Root Mean Square Error. From the obtained results, it was possible to verify that as the density of GPS trajectories and the resolution of the frequency image increase, the accuracy of the extracted road axes gets better.