Neural differential equations based diffusion models for high-jump posture prediction

To address the challenges faced by traditional high jump posture prediction methods -namely environmental noise and motion continuity -this paper proposes an innovative solution integrating diffusion models with neural differential equations.By leveraging diffusion models to learn the true posture distribution from noisy data and employing neural differential equations to capture continuous motion dynamics, the approach effectively overcomes the limitations of conventional methods in modelling complex movements.Experiments on the human3.6mand dynamically informed pose from inertial measurement unit public datasets demonstrate that this method achieves a prediction accuracy of 96.2%, surpassing mainstream long short-term memory models by 8.7%, while reducing joint angle error to 3.2 degrees.These findings provide more reliable technical support for sports science training and sports injury prevention.