Robust rotation estimation from three orthogonally oriented cameras

This thesis develops the ORThogonally-oriented CAMeras (ORTCAM) algorithm for robustly estimating rotation of a set of three orthogonally-oriented cameras undergoing motion. Orthogonally-oriented cameras are any set of diverging cameras where each camera is oriented 90° to all other cameras. The ORTCAM algorithm robustly estimates rotation using feature correspondences between two views from each camera. The ORTCAM algorithm is shown to both resolve planar and rotation degeneracies and yield rotation estimates superior to the those obtained by averaging the rotation estimates from all cameras. The assumptions made in the algorithm are that the cameras are exactly oriented 90° apart, the images have been calibrated and the maximum percentage of outliers between two views is 50%. However, simulations show that the ORTCAM algorithm is robust to small deviations in orthogonality of the cameras.