Multi-beam terrain/object classification for underwater navigation correction

Building accurate bathymetries of the seabed has been a focus of study in the last decade. For this purpose seabed point cloud registration has been a focus for some researchers. Some of this registration methods are based on gathering the points of the cloud that contain more information for the registration (i.e. the ones that flat or smooth, normally being the seabed) and using them as part of ICP-derived methods. For this point picking purpose, we present a segmentation technique that distinguish between objects (interesting for registration) and ground (smooth and not interesting for registration). The method proposed here uses difference of normals for object's border detection and a variation of the Density-Based Spatial Clustering of Application with Noise for object clustering. Once the objects boundaries are detected and the points are clustered the rest of the points are classified as object or ground. This classification is done by taking all the points that lie within the object's border and checking it's depth compare to its closes border point. The method is evaluated using a multi-beam dataset gathered on the La Lune shipwreck, a site of archaeological interest.