Assessment of Motion Activity for a Rainbow Trout Flock in Underwater Video Surveillance System

Video Surveillance Systems (VSS) are progressing towards the digitalization of industry. In this demo, we apply the Artificial Intelligence (AI) technology to assess the motion activity of a school of rainbow trout. The practical need is to detect and react on too low and too high activity. The input data come from stereo cameras installed around the pool. The activity recognition used the specifically trained YOLO-Pose neural network (NN). Our first algorithm assesses the speed of a rainbow trout based on averaging the individual fish activity parameters. Our second algorithm assesses the angle of an individual rainbow trout to detect the fish in bad health state. Our early experimental study of the algorithms demonstrates their applicability for monitoring the fish health and condition, feeding management, water quality control and fish behavior control in aquaculture