Image mosaicking is one of the most dynamic areas in underwater research. In particular, 3D mosaicking is an important tool for accurate mor-phometric measurements and seabed exploration that improves visualization and navigation underwater. The automatic definition of 3D mosaics involves two main steps. In the first, depth maps are evaluated from a stereo vision system, where the latter frames from different angles are registered. Depth maps require stereo disparity evaluation, which deals with the mathematically ill-posed stereo correspondence problem due to occlusions, camouflage, untextured surfaces and so forth. The disparity is handled in a reliable way, with local and global matching methods and without user intervention. 3D registration is accomplished by tracking reliable features in adjacent stereo pairs. Experimental results validate the suggested algorithmic framework.
展开▼
