Evaluating bias in stereo camera measurements using swimming behavior simulations

Abstract

The use of stereo camera systems to measure fish body size has gained attention in aquaculture systems management. However, the different correlations between swimming depth and size according to fish species can lead to measurement bias depending on camera range and position. This study accordingly applied a fish behavior model based on the Boid model to simulate the relationship among camera position, viewing range, and measurement results considering swimming depth behavior. A total of 722 simulated fish (yellowtail Seriola quinqueradiata) were placed into a rectangular aquaculture net pen model. Virtual cameras were located at different depths along the side, above, and below the pen to measure the total length of individual fish within view. Horizontally oriented cameras closer to the surface captured significantly larger/smaller individuals when the behavior model assumed that larger individuals swim at shallower/deeper depths; vertically oriented cameras facing down from the water surface or up from the bottom net provided mean total length measurements might close to the actual mean total length of the population when a wide viewing range was provided. The proposed simulation method can be applied to preliminarily evaluate the measurement biases inherent to fish behavior and thereby minimize their influence on camera measurements.