Automatic Generation of Road Maps from Low Quality GPS Trajectory Data via Structure Learning

Up-to-date road maps are critical for intelligent transportation applications, such as car navigation and route planning. Unfortunately, conventional road map generation through surveying and image digitizing is labor intensive and time consuming. Nowadays, a large volume of GPS trajectory data can be widely acquired from vehicles mounted GPS receivers. These GPS trajectories are natural data resources for generating up-to-date road maps. Currently, extensive road map generation methods have been developed using the GPS trajectories, however, duo to the noise, low sampling rate, and uneven density distribution of GPS trajectories, most of the state-of-the art methods are not robust enough and usually sensitive to noises. Furthermore, extensive preprocessing and complex refinement steps reduce the power and flexibility of these methods. To produce a more robust and flexible road map generation method from GPS trajectories, a novel method is proposed in this paper. The main innovation part of this paper is that we generate a road map from massive GPS trajectories through principal graph structure learning and tree linking strategy, in this manner, a best-fitting graph structure (principal graph) of GPS points is constructed by optimizing an objective function. Using this method, the influences of noise, low sampling rate, and uneven density distribution can be minimized, which makes the proposed method more robust. The simulated datasets and the real-world GPS trajectory data were used in experimental analysis. The experimental results show the robustness, validity, and strength of the proposed method.