Abstract
Camera estimation of fish size in aquaculture production and fisheries is crucial for enabling knowledge-based decision making. Automatic estimation of fish lengths on an individual and population basis are key steps to automate biomass estimation. We propose to use an underwater 3D range-gated camera for accurate fish-length estimation of free-swimming fish. The proposed algorithm requires no manual work or contact with the fish and is done in-situ. A robust algorithmic pipeline consisting of detection, tracking and fish length estimation stages is proposed. We show the accuracy of the proposed system on free-swimming fish, both in terms of individual fish lengths as well as the population distributions. The results show that the proposed system achieves length estimation errors in the order of 1% of manual-measured fish length.