A Method of Extracting Road Network Structure from Trajectory Data Based on U-Net Network

Abstract

Accurate road network plays a very important role in urban traffic application. The traditional methods to generate road network are expensive and the update of the road network is not timely enough. With the widespread use of Global Positioning System (GPS) embedded equipment, lots of moving objects can generate a large amount of trajectory data, from which it could become possible to extract road network information. The existing road network extraction methods require different prior experiences and parameters for road networks in different regions, and the effect is not satisfactory. In this article, we propose a method to generate city road network structure based on an improved U-Net network, which can extract road network from trajectory data. More specifically, we first learn the existing road network structure and extract the feature from trajectory data, then use the improved U-Net network to infer the road centerline, finally, we postprocess its topology and generate the final road network. Our method has been validated on different trajectory datasets and achieved good visualization results.