Author(s): Jeffrey Tuhtan; Elizaveta Dubrovinskaya; Lizaveta Miasayedava; Vishwajeet Pattanaik; Jurgen Soom; Bernd Mockenhaupt; Cornelia Schutz; Christian Haas And Philipp Thumser
Linked Author(s): Jeffrey Tuhtan
Keywords: No keywords
Abstract: Here we present a new multi-stage computer vision system for automatically processing underwater fish counter videos. The camera-agnostic system consists of four separate software modules. Module 1 classifies seven typical environmental conditions: clear water, turbidity, debris, aerated flows, low-lighting, lighting overexposure and biofouling. Module 2 then automatically sorts fish and no-fish videos. Module 3 will rely on convolutional neural networks to detect fish species, family and size class in 5 cm increments. Finally, Module 4 will track individual fish during their passage through the counter to assess migration direction relative to the flow, and distinguishes between fish which remain for long periods inside of the counter, those which migrate up or downstream through the counter, and those which enter and exit without migrating through the fish counter.
Year: 2022