Frequency-based underwater terrain segmentation
Bertrand Douillard · Navid Nourani-Vatani · Matthew Johnson-Roberson · Oscar R. Pizarro · Stefan B. Williams · Charles T. Roman · Ian Vaughn
TL;DR
Treating seafloor elevation maps as signals, this method separates ground from objects by frequency cut-off, producing segments useful for navigation across varied terrains.
BRIEF
Underwater robots build 3D maps of the seafloor, but separating objects such as rocks or coral from the ground beneath them is difficult when the terrain is uneven. This work converts 3D point clouds into 2D elevation maps and examines them in the frequency domain, where slowly varying undulations correspond to the ground. A cut-off frequency, found automatically by peak detection or guided by a maximum object size, defines the ground surface. Points above it are clustered into objects. Tests against hand-labelled data and sub-map alignment show it works across varied terrains.
Abstract
A method for segmenting three-dimensional data of underwater unstructured terrains is presented. The three-dimensional point clouds are converted to two-dimensional elevation maps and analyzed for segmentation in the frequency domain. The lower frequency components represent the slower varying undulations of the underlying ground. The cut-off frequency, below which the frequency components form the ground surface, is determined automatically using peak detection. The user can also specify a maximum admissible size of objects to drive the automatic detection of the cut-off frequency. The points above the estimated ground surface are clustered via standard proximity clustering to form object segments. The precision of the segmentation is compared against ground truth hand labelled data acquired by a stereo camera pair and a structured light sensor. It is also evaluated for registration error when the extracted segments are used for sub-map alignment. The proposed approach is compared to three point cloud based and two image based segmentation algorithms. The results show that the approach is applicable to a range of different terrains and is able to generate features useful for navigation.