Online generation of an underwater photo map with improved Fourier Mellin based registration

A novel algorithm for stitching images, which is faster and more robust than standard approaches like the Scale Invariant Feature Transform (SIFT) is presented. The algorithm is particularly suited for Autonomous Underwater Vehicles (AUV), namely for the online generation of photo maps that can be the basis of intelligent onboard functionalities. The photo maps can be generated just based on registration, i.e., without any information about the vehicle's pose or its motion. A new method based on the Fourier-Mellin (FM) transform is presented. It is based on a 2D Fourier transform where the shift between two signals is detected by the phase information. In contrast to previous work using FM, a polar-logarithmic resampling of image information is used to turn rotation and scaling into corresponding phase shift and allow for a registration in one step.