Automatic mosaic creation of the ocean floor

We describe the automatic creation of video mosaics of the ocean floor, which deals with the problem of image motion estimation in a robust and automatic way. The motion estimation presented in this work is based on an initial matching of corresponding areas over pairs of images. As the error prone nature of the matching process is a commonly overlooked problem, this paper makes use of robust matching techniques, which can cope with a high percentage of wrong matches. In our approach, several motion models are established under the projective geometry framework, allowing the creation of high quality mosaics where no assumptions are made on the camera motion. This as an improvement over traditional approaches for underwater mosaicing, usually relying on the camera to be facing the sea floor, so that the image plane is approximately parallel to the floor plane. Extensive tests were run on underwater image sequences, testifying the good performance of the implemented matching and registration methods. Even with notorious violations of the underlying assumption of static planar scenes, the algorithm can still find the motion parameters as to create mosaics with bearably noticeable misalignments to the human eye