Implementation of a Fish Size Measurement System Using a Monocular Camera

Abstract

Recently, fish farming's have been increasing to keep food production. Fish body size is one of the most important factors to determine an appropriate feeding rate. Thus, this data is often measured manually by directly catching fish with a net. However, this measurement method needs time-consuming, and hard work and leads damage and stress to fish. For overcoming this problem, an automatic measurement method with machine learning by using monocular camera is proposed in this paper. In general, conventional measurement methods are based on a stereo camera, which needs often large hardware amount and expensive. On the other hand, the proposed method is based on a monocular camera, which is small hardware amount and less expensive, this system can get depth data by utilizing machine learning algorithm and to automatically detect fish in images which are captured by a monocular camera. From experimental results on actual aquarium with dorado fish. The average relative error of the measurements is about 12.4%, which is a relatively small error to obtain the fish size.