Coastal marine ecosystems are rapidly changing, driven by multiple stressors ranging from ocean warming, fisheries, invasive species, habitat loss and degradation. Mitigating these effects requires effective observation methods to first inform the policy and management options, and then to monitor their effectiveness. Underwater cameras now offer a cost-efficient and attractive alternative to the labour-intensive fishing gears traditionally used in surveys. Large-scale deployment would produce unprecedented volumes of observations, yet only a fraction of these images can be analysed manually. Advances in machine learning provides the opportunity for developing fully automated video analysis, but to reach the point where deep learning-based computer vision can seamlessly be incorporated into field surveys, significant effort needs to be allocated to building high quality annotated datasets and to explore methods that require less data for training and verification.
ShareVision will explore such methods with aim of developing computer vision systems that will improve our abilities to monitor, understand and manage coastal fisheries. We will assemble and build video and image library for coastal species in Norway (and elsewhere) to be used to develop robust tools for detection, classification and tracking of fish in natural environments. A key element for achieving successful results is our access to high-end equipment and big data through our partner Institute of Marine Research.
ShareVision is collaborating with the research projects CoastVision and ARVEN.