Related Experiment Video
Updated: Jul 4, 2026
06:52
Deep-Learning Based Multi-Joint Synchronous Tracking for Objective Quantification of Hindlimb Locomotor Kinematics in Rats
Published on: April 3, 2026
Motor ability-aware adaptive pose estimation with hierarchical uncertainty modeling and cross-ability learning.
Qian Wang1, Quan Zhou2, Jinshan Yang3
1College of Physical Education, Hunan University of Science and Technology, Changsha, Hunan, China.
Scientific Reports
|July 2, 2026
Summary
This study introduces a novel human pose estimation framework that adapts to diverse motor abilities, significantly improving accuracy and reducing performance gaps for individuals with motor impairments. The system enhances inclusivity in assistive technologies.
Area of Science:
- Computer Vision
- Machine Learning
- Biomedical Engineering
Background:
- Current human pose estimation (HPE) systems exhibit reduced accuracy with varying motor abilities.
- Existing HPE methods lack uncertainty quantification due to reliance on standardized anatomy.
Purpose of the Study:
- To develop an inclusive HPE framework addressing diverse motor abilities.
- To quantify uncertainty in pose estimation across different body parts and user groups.
Main Methods:
- MADSRNet: Dynamically adapts skeleton structures using gated fusion and dynamic graph construction.
- HUGPose: Employs heteroscedastic regression for hierarchical uncertainty estimation.
- CADCL: Leverages contrastive learning and domain adversarial training for cross-ability knowledge transfer.
Main Results:
- Achieved state-of-the-art results with 41.2 mm MPJPE on the DiverseMotor-PE dataset, a 14.7% improvement over ViTPose.
- Reduced the mean per-level MPJPE difference between typical and severely impaired individuals from 12.1 mm to 10.3 mm.
- Demonstrated low expected calibration error (1.58%) for uncertainty estimates.
Conclusions:
- The proposed framework provides accurate human pose estimation irrespective of motor ability.
- This work advances inclusive HPE systems, promoting equitable access to healthcare and assistive technologies.
