A Fast and Robust Large-Scale Structure from Motion Using Auxiliary Information

It's an effective mean for large-scale 3D scene reconstruction using a global structure from motion (SfM) pipeline. However, those methods are time-consuming in common. The auxiliary information from the EXIF is helpful for the reconstruction process. In the paper, we will use the GPS data to select potential matching images, and some distant image pairs will be discarded directly without calculating feature matching. This strategy can greatly accelerate the reconstruction process. We also use two different manners to deal with the feature correspondences in order to speed up the process and get more points in the final structure. When estimating the global camera poses, we add a global residual item to the optimization equation to make the camera poses more robust. Our method performs fast and robust on several image sets of real scene captured by Unmanned Aerial Vehicle.