Wild Motion Unleashed: Markerless 3D Kinematics and Force Estimation in Cheetahs
Authors:
Zico da Silva,
Stacy Shield,
Penny E. Hudson,
Alan M. Wilson,
Fred Nicolls,
Amir Patel
Abstract:
The complex dynamics of animal manoeuvrability in the wild is extremely challenging to study. The cheetah ($\textit{Acinonyx jubatus}$) is a perfect example: despite great interest in its unmatched speed and manoeuvrability, obtaining complete whole-body motion data from these animals remains an unsolved problem. This is especially difficult in wild cheetahs, where it is essential that the methods…
▽ More
The complex dynamics of animal manoeuvrability in the wild is extremely challenging to study. The cheetah ($\textit{Acinonyx jubatus}$) is a perfect example: despite great interest in its unmatched speed and manoeuvrability, obtaining complete whole-body motion data from these animals remains an unsolved problem. This is especially difficult in wild cheetahs, where it is essential that the methods used are remote and do not constrain the animal's motion. In this work, we use data obtained from cheetahs in the wild to present a trajectory optimisation approach for estimating the 3D kinematics and joint torques of subjects remotely. We call this approach kinetic full trajectory estimation (K-FTE). We validate the method on a dataset comprising synchronised video and force plate data. We are able to reconstruct the 3D kinematics with an average reprojection error of 17.69 pixels (62.94 $\%$ PCK using the nose-to-eye(s) length segment as a threshold), while the estimates produce an average root-mean-square error of 171.3 N ($\approx$ 17.16 $\%$ of peak force during stride) for the estimated ground reaction force when compared against the force plate data. While the joint torques cannot be directly validated against ground truth data, as no such data is available for cheetahs, the estimated torques agree with previous studies of quadrupeds in controlled settings. These results will enable deeper insight into the study of animal locomotion in a more natural environment for both biologists and roboticists.
△ Less
Submitted 10 December, 2023;
originally announced December 2023.
AcinoSet: A 3D Pose Estimation Dataset and Baseline Models for Cheetahs in the Wild
Authors:
Daniel Joska,
Liam Clark,
Naoya Muramatsu,
Ricardo Jericevich,
Fred Nicolls,
Alexander Mathis,
Mackenzie W. Mathis,
Amir Patel
Abstract:
Animals are capable of extreme agility, yet understanding their complex dynamics, which have ecological, biomechanical and evolutionary implications, remains challenging. Being able to study this incredible agility will be critical for the development of next-generation autonomous legged robots. In particular, the cheetah (acinonyx jubatus) is supremely fast and maneuverable, yet quantifying its w…
▽ More
Animals are capable of extreme agility, yet understanding their complex dynamics, which have ecological, biomechanical and evolutionary implications, remains challenging. Being able to study this incredible agility will be critical for the development of next-generation autonomous legged robots. In particular, the cheetah (acinonyx jubatus) is supremely fast and maneuverable, yet quantifying its whole-body 3D kinematic data during locomotion in the wild remains a challenge, even with new deep learning-based methods. In this work we present an extensive dataset of free-running cheetahs in the wild, called AcinoSet, that contains 119,490 frames of multi-view synchronized high-speed video footage, camera calibration files and 7,588 human-annotated frames. We utilize markerless animal pose estimation to provide 2D keypoints. Then, we use three methods that serve as strong baselines for 3D pose estimation tool development: traditional sparse bundle adjustment, an Extended Kalman Filter, and a trajectory optimization-based method we call Full Trajectory Estimation. The resulting 3D trajectories, human-checked 3D ground truth, and an interactive tool to inspect the data is also provided. We believe this dataset will be useful for a diverse range of fields such as ecology, neuroscience, robotics, biomechanics as well as computer vision.
△ Less
Submitted 24 March, 2021;
originally announced March 2021.
