A Stereo Vision Framework for 3-D Underwater Mosaicking

Research on automatic mosaic creation for underwater applications has been investigated in the last fifteen years. The reconstruction of complex 3-D structures is useful in several underwater applications; in particular, 3-D mosaicking constitutes an important tool for seabed exploration, improving visualization and navigation in the underwater medium. Moreover, underwater measurement systems have been used extensively in marine research to estimate the size of interesting objects such as organisms and structures. In addition, visual sensing can be an enabling technology for Autonomous Underwater Vehicles (AUVs), which have the critical requirement to maintain an ongoing representation of its relative position with respect to an environmental representation. In these contexts, a better perception of the underwater environment can be achieved by using image processing algorithms for a suitable representation of the seabed. In optic sensing field and underwater environment, shape acquisition concerns with sensing activity including Shape from Stereopsis, Shape from Photometric Stereo, Shape from Motion and Active Stereo. Main aspects about Shape from Stereopsis will be presented focusing the attention on synchronism in the acquisition at hardware/software levels. The usage of triggered expensive equipments in synchronized stereo sequence acquisitions could present some technical limitations, such as low frame rate and poor images resolution, demoting the quality of the 3-D mosaic. For the previous reasons, an ad-hoc algorithmic solution will be detailed to limit asynchronism problems due to time gap (delay) in stereo frames acquisition, when low-cost non-professional (non-triggered) hardware is used. To achieve a metric reconstruction, the presented framework requires a calibration phase whereas a new stereo frame-selection scheme based on the Epipolar Gap Evaluation (EGE) will be discussed to overcome asynchronisms. Normally, dense depth maps obtained by evaluating dense disparity maps allow the construction of high quality 3-D mosaics. A detailed discussion about methodologies for dense stereo matching will be addressed to handle the ill-posed stereo correspondence problem. A joined use of local (respectively global) matching methods and imaging enhancement algorithms will be considered to identify an appropriate amount of right correspondences in severe underwater conditions (backscattering, brightness constancy violation condition). The chapter treats an important aspect of the 3-D mosaic reconstruction as the registration by ego-motion and camera pose estimation. Classical solutions (i.e. Iterative Closest Point algorithm) and new trend in the O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m