Fish Tracking by Machine Vision for Indoor Farming Systems

Abstract

Singapore is exploring the development of multiple-story buildings for farming to provide more spaces for fish farming and open up more opportunities for shared facilities and resources to relieve the limited space and the manpower intensive requirements. Therefore, it is necessary to apply new technologies on production techniques to optimize fish farming operations, equipment and infrastructures for efficiency and accuracy. Video tracking is one of major approaches to analyse high temporal and spatial resolutions of animals in groups. In this work on indoor farming system of fishes, video capturing approaches and video unmarking approaches are investigated and compared to attain suitable solutions to understand physical characteristics of fishes and environmental factors to best inform successful fish farming. Video background removal and video fish segment approaches are proposed to analyse the trajectories of fish movement such as fish speed, swim patterns etc. To simulate the indoor farming environment, a fish tank with a fish physical characteristics monitoring system is established. The collected data are stored in dedicated workstations. The collected data are stored in dedicated workstations. The system employs conventional cameras with output of grayscale images. The system is fully automatic and can achieve near to high correct identities.

Keywords: Machine Learning Fish Tracking Fish Behaviour