Fully Automatic UAV Image-Based Sensor Orientation

The paper presents a procedure for the automatic orientation of image blocks captured by UAV-borne cameras. Disregarding the availability or absence of GPS/INS data, the method is able to extract a set of manifold tie points to be used as observations in a photogrammetric bundle adjustment. Complex image configurations and scenes can be dealt with, encompassing the use of convergent imagery, strong perspective deformations, lighting changes and so on. Tie points are extracted among image pair combinations by exploiting operators like SIFT and SURF, which are invariant with respect to scale variations and rotations as well. Then all image pairs are linked together to compute a global bundle adjustment. In order to speed up the process, a coarse-to-fine approach can be used, integrated to the creation of a visibility map that limits the selection of image pairs to be analysed. Some tests on different UAV blocks are reported and discussed, showing the potential of this procedure to obtain the orientation of all images in a fully automated way.