Vision-based motion sensing for underwater navigation and mosaicing of ocean floor images

Abstract

The authors propose a direct solution for real-time estimation of the 3D motion of a submersible vehicle from the video images of the seafloor, acquired by an on-board camera, in order to perform the following tasks: Maintaining a prescribed trajectory, or determining position relative to a reference point, during short-range navigation along the seafloor. Constructing a mosaic image of the bottom scene within the scanned area by registering consecutive frames. The authors present experimental results with images recorded in a water tank to evaluate performance based on accuracy in the estimation of 3D motion and the registration of consecutive frames in the mosaic image.