Generating three-dimensional skeleton representations of aquatic animals using machine learning

A system for generating a three-dimensional skeleton representation of an aquatic animal among a population of aquatic animals in an aquatic environment, comprising:

• a device with two open ends connected by a channel and configured to receive water from the aquatic environment through the channel;
• one or more cameras attached to the walls of the device and directed towards the interior of the device;
• at least one processor configured to receive image data from the one or more cameras and to process the image data to:
• identify key points on or inside the aquatic animal, including occluded key points, and their locations in the one or more 2D images;
• generate one or more 2D skeleton graphs represented as nodes connected by edges, wherein each node in a 2D skeleton corresponds to one identified key point;
• calculate estimated 3D positions for nodes in the one or more 2D skeleton graphs;
• determine the 3D coordinates of the nodes of a 3D skeleton from the estimated 3D positions of the nodes of the one or more 2D skeletons, and generate 3D skeleton graphs as a pre-defined structure of nodes with 3D coordinate positions, connected by edges representing straight line distances between respective pairs of nodes;
• store the 3D skeleton in a database as a representation of the individual aquatic animal or transmit the 3D skeleton as a data structure including the structure of nodes connected by edges;
• estimate the biomass of the aquatic animal by performing mathematical operations on one or more of the nodes and one or more of the straight line distances in the stored or transmitted 3D skeleton graph;
• and
• track growth in the population based on biomasses estimated from respective 3D skeleton graphs generated at different times.