AI for automated shark tracking and biometrics from aerial imagery

In collaboration with Dr. Mario Espinoza and Sergio Madrigal-Mora from the University of Costa Rica, we developed an AI tool (FLAIR) which detects, segments, and extracts biometric information from drone imagery of animals. We applied this tool on a dataset of Pacific Nurse Sharks in Santa Elena Bay, Costa Rica and found it performs with comparable accuracy to a human annotator for measuring key biometrics such as body length and tailbeat frequency. Notably, this method requires no training data or machine learning expertise to use. By simply changing the prompt, we found FLAIR succeeded on videos of white sharks, blacktip reef sharks, zebras, and even birds. 

​

Find the full paper in Methods in Ecology and Evolution, the prior work on standard object detectors in the Proceedings of the European Conference on Computer Vision, the plain-language blog explaining our method, a highlight from Save Our Seas Foundation, and a short video by Stanford’s Woods Institute for the Environment on our fieldwork.