WhaleTrack: Telemetry-Driven Georeferenced Multi-Whale Tracking from Drone Video
Sep 8, 2026·,
,,,,,,,,,,,·
1 min read
Pierre Adorni
Lucie Laporte-Devylder
Matthieu Le Lain
Thomas Fillon
Louis François
Kshitij Raj Sharma
Shivam Pande
Manuel Nkegoum
Sergio Suzerain Osson
Emanuel Goulart Farias
Frédéric Raimbault
Charlotte Pelletier
Marc Chaumont
Sébastien Lefèvre
Abstract
WhaleTrack is a modular system for tracking humpback whales from drone video. It combines oriented bounding-box detection, multi-object tracking with lightweight spatial re-linking, and telemetry-driven ray-plane reprojection using the drone GPS, altitude and gimbal orientation, together with a per-frame field-of-view estimation. The system outputs georeferenced trajectories together with the instantaneous observation footprint, which helps interpret non-detections in moving-camera footage. The detector reaches 94.7% mAP50 on a held-out flight, and length estimates against a 7.3 m reference vessel show a mean signed error of -0.44 m. The dataset, model weights and pipeline are released to support reproducible drone-based marine-mammal monitoring.
Type
Publication
2nd Workshop on Marine Vision, in conjunction with ECCV 2026
Co-authored paper presented at the 2nd Workshop on Marine Vision, held in conjunction with ECCV 2026 in Malmö, Sweden, on September 8, 2026.
A side project outside astronomy, within the IRISA OBELIX team: the same computer vision toolbox, applied to humpback whale monitoring from drone footage collected off Los Cabos, Mexico, in January 2026.
Authors
AI for astronomy & astrophysics
PhD student at IRISA, Université Bretagne Sud (expected 2026), and lecturer in computer science,
working on foundation models for astronomy and astrophysics, with a broader interest
in deep learning applied to scientific data.
Also contributing to the UniverseTBD collaboration on vision-language models for astronomy.