3D Pose Estimation for Seabirds

Content / Page
add_circle_outline 2022-10-24
Short description of the topic: Seabirds exhibit a range of distinct behaviors on the cliffs, related to e.g. socialization, thermoregulation and stress. Developing methods to capture and quantify these behaviors could help researchers better understanding causes behind behavior that is ultimately determining the success of these birds in raising chicks. As a step to develop behavioral classification, this project aims at building a 3D model of the birds’ postures, based on footage from multiple and monocular cameras. Annotations (~650 images) and a trained model for 2-dimensional pose estimation (DeepLabCut) is at disposal for the project.

While 3D pose estimation for humans is well studied, for birds it is a largely unexplored research field with its own challenges, such as strong changes of the bird’s volume, highly articulated wings and feathers, and limited labeled training data. The goal of this project is to explore the transferability of recent advances in 3D human pose estimation to the existing dataset of seabirds.

Your organization: Linköping University, Computer Vision Laboratory, Swedish Agricultural University

Contact person: Bastian Wandt, bastian.wandt@liu.se

Preferred student background: Knowledge of Machine Learning and Computer Vision 

Preferred location: Location is Linköping University, Computer Vision Laboratory.

Attributes

Master Thesis Proposals