Title:	Horse Shape and Pose Capture
DNr:	Berzelius-2023-332
Project Type:	LiU Berzelius
Principal Investigator:	Hedvig Kjellström <hedvig@kth.se>
Affiliation:	Kungliga Tekniska högskolan
Duration:	2023-12-03 – 2024-07-01
Classification:	10207
Homepage:	https://github.com/Celiali/hSMAL
Keywords:

Abstract

Horses are the most valuable domestic animal, and a large industry focuses on their breeding, care, training, and use in sports. Owning and keeping a horse is a large investment of money, time, and resources. Horses are delicate although they are large animals, because of their skeletal and tendinous limb structures. Consequently, horses are widely studied from a behavioral and biomechanical perspective, to evaluate their performance, prevent injuries, and ensure their health, welfare, and performance. One of the promising approaches to enhance this understanding is through 3D reconstruction, which creates detailed and accurate 3D horse bodies and movements. It’s challenging since 3D data collection needs specific equipment and modeling motion is complex. While this field is still under-exploring. Our task is to develop a deep learning and computer vision technique for capturing 3D horse reconstruction from multi-modal data, like images, mocap data, or audio. We train and validate our techniques using existing datasets of horses, as well as new datasets collected from various sources, like farms, or online. This will be the continuation of student Ci’s project berzelius-2023-256 which will terminate early on 2023-12-01 when this project starts.