Computer Vision Based Intelligence for Tracking and Predicting Whale Blow Trajectories

Collecting blow samples from whales allows for studies on individual and population health. However, it is important to not interfere or harm the whales, forcing collections to happen from a distance. An Uncrewed Aerial Vehicle (UAV) equipped with a collection system is capable of collecting whale blow samples without interfering with the whale by flying through the blow cloud. The work presented describes a computer vision based intelligence system for facilitating whale blow collection using an UAV. The computer vision system utilizing Ultralytics Neural Network You Only Look Once (YOLO) version 8. Using the timing of YOLO's outputs, an averaging technique is used to predict future blow timings from past ones. Additionally, an Auto Regressive Moving Average (ARMA) model is used in combination with the bounding boxes provided by YOLO in order to predict where the blow centres will be in the near future. It is shown that YOLO is able to successfully identify a whale, its head, and the blow while the ARMA model is able to predict where the blow will be sufficient accuracy for collection.