Automated Observation of Spatial Distributions of Reef Fauna

Abstract

Advances in underwater imaging equipment and autonomous underwater survey platforms have made it possible to routinely collect large numbers of underwater images and generate georeferenced 3D reconstructions of reef-scale seafloor extents. The scale of the datasets generated by these technologies present an opportunity to broaden the scope of our understanding of benthic habitats.

Two barriers to the use of image-based habitat reconstructions for routine observation of benthic ecosystems are addressed. Firstly, the high volume of image data produced by these technologies require time-consuming manual annotation for processing. This has, to date, made them infeasible for regular observation of benthic habitats. Secondly, mobile fauna are typically not observed in these surveys since their movement can cause them to be missed, duplicated, or distorted in the reconstructions. This work addresses these challenges using automated image analysis and exploiting the navigation information and georeferencing contained in these data to detect and localise benthic targets at reef scales.

The key contributions are a unique approach to observing the spatial distributions of benthic targets from moving imaging platforms, implementation of automated object detection to estimate the fine scale distributions of mobile benthic targets from reef-scale photomosaics using minimal manually annotated imagery, and a novel segmentation approach for diffuse benthic targets, integrating the appearance and structure information from image-based reconstructions.

The methods developed in this work allow efficient mapping of habitat features over large spatial extents, facilitating detailed analysis of the interactions between and among these inhabitants and the structure of their environment. Being able to efficiently observe these fine scale distributions over entire reefs provides a unique way of observing benthic ecosystems which is not possible using traditional observational techniques.