Rectification of 3D building models based on GPS signal collected by vehicle

For autonomous driving, both the estimation of the accurate ego position of vehicle and creation of the environment map such as Simultaneous localization and mapping (SLAM) technology are essential. In the SLAM technology, the 3D building model becomes an important aid to many positioning methods such as LiDAR and GPS positioning methods. To build accurate 3D building models, the accurate building footprint (the boundary of the building) is required. In this study, we propose an innovative method to correct the errors of building footprint on the 3D map by using GPS signal. In the urban canyon, GPS signal will be blocked by the buildings and only its reflection signal is received, which is well-known as non-line-of-sight (NLOS) reception. These reflections are potentially capable of indicating the correct position of the buildings. By using of a rough 3D building model, we apply it with a GPS ray-tracing method to track the simulated reflection path of the NLOS GPS. Theoretically, the length of observed reflection path, which is well-known as pseudorange measurement, and the length of simulated reflection path should be very similar. However, if the 3D map is not accurate, the difference between the observed pseudorange and simulated pseudorange will be detected. To utilize this fact, the proposed method is able to estimate the true position of the wall on the 3D map. The experiment results show that the proposed method successfully corrected the footprint of the rough 3D building model into about 1 meter accuracy. Importantly, the proposed method is capable of rectifying the building model only if several reflections GPS signals can be received from a same building.