DeepVision: a stereo camera system provides highly accurate counts and lengths of fish passing inside a trawl

Abstract

The DeepVision stereo camera system collects a continuous record of colour images of all fish passing inside the extension of a trawl. Ninety-eight percent of 1729 fish captured while trawling could be identified to species from images, and lengths could be estimated from the images of 96% of the fish identified. A landmark distance technique developed to estimate lengths from images containing incomplete, curved, or obscured fish introduced <1% error for the majority of individuals (maximum 5% error). The technology can greatly increase the scope of information collected during trawl sampling, including documenting fine-scale distribution of individual fish and species overlap. Such information can aid in interpreting acoustic data and fine-scale community composition and could be collected with an open codend trawl, greatly reducing sampling mortality. Images are easily archived, providing an opportunity to quality check the raw data and revisit datasets originally collected for different purposes. Adaptation of the technology for commercial fisheries could reduce the catch of unwanted species and sizes.