Heterogeneous Data Registration for 3D Underwater Scene Reconstruction

This paper addresses the heterogeneous data registration problem, which is one of the key features for any scene reconstruction and representation, especially for the underwater environment. In this study, we propose a registration method built around a 2D-to-3D feature-based approach that registers high-resolution side-scan sonar images with bathymetric data (topographic 3D point cloud) obtained by multibeam echosounder. This process enables us to achieve a global 3D mosaic of the studied underwater scene, which is informatively richer and more reliable than each individual dataset. Indeed, the interest of this data fusion representation is that it combines the benefits of using each sensor: bathymetric information provides the geometric structure of the sea-bottom, while sidescan sonar images contribute a complementary observation with better resolution of the sea-bottom reality (e.g. sedimentology, bottom-laying object, etc.)