The development of maritime applications require monitoring, studying and preserving of detailed and close observation on the underwater seafloor and objects. Stereo vision offers advanced technologies to build 3D models from 2D still overlapping images in a relatively inexpensive way. However, while image stereo matching is a necessary step in 3D reconstruction procedure, even the most robust dense matching techniques are not guaranteed to work for underwater images due to the challenging aquatic environment. In this thesis, in addition to a detailed introduction and research on the key components of building 3D models from optic images, a robust modified quasi-dense matching algorithm based on correspondence propagation and adaptive least square matching for underwater images is proposed and applied to some typical underwater image datasets. The experiments demonstrate the robustness and good performance of the proposed matching approach.
Center for Coastal and Ocean Mapping
Hu, Han, "Euclidean reconstruction of natural underwater scenes using optic imagery sequence" (2015). Center for Coastal and Ocean Mapping. 715.