Towards Generalized Benthic Species Recognition and Quantification Using Computer Vision

Seabed resource exploitation and conservation efforts are extending to offshore areas where the distribution of benthic epifauna (animals living on the seafloor) is unknown. There is a need to survey these areas to determine how biodiversity is distributed spatially and to evaluate and monitor ecosystem states. Seafloor imagery, collected by underwater vehicles, offer a means for large-scale characterization of benthic communities. A single submersible dive can image thousands of square metres of seabed using video and digital still cameras. As manual, human-based analysis lacks large-scale feasibility, there is a need to develop efficient and rapid techniques for automatically extracting biological information from this raw imagery. To meet this need, underwater computer vision algorithms are being developed for the automatic recognition and quantification of benthic organisms. Focusing on intelligent analysis of distinct local image features, the work has the potential to overcome the unique challenges associated with visually interpreting benthic communities. The current incarnation of the system is a significant step towards generalized benthic species mapping, and its feature-based nature offers several advantages over existing technology.