Fast Neural Inverse Kinematics on Human Body Motions

2025-06-22

AI Summary

To address the slow inference speed and limited real-time applicability of markerless motion capture, this paper proposes a lightweight and efficient neural inverse kinematics framework that directly regresses joint angles from single-frame 3D keypoints. Methodologically, we design a real-time-optimized neural architecture integrating channel-spatial collaborative compression and hierarchical feature reweighting, coupled with a joint-constraint-aware loss function and a progressive training strategy. Ablation studies systematically validate the efficacy of each component. Quantitative evaluation across multiple benchmark datasets demonstrates state-of-the-art accuracy (mean angular error < 8.5°) alongside an inference speed of 210 FPS on an RTX 4090—3–5× faster than existing methods. Qualitative results further confirm robustness and generalization under complex motions. Overall, this framework significantly enhances the practicality and deployability of markerless motion capture.

Technology Category

Computer Vision: Motion & TrackingMachine Learning: Imitation Learning & Inverse Reinforcement LearningCognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

Responsible Web: Machine-in-the-loop, human agency and autonomySystems and Infrastructure for Web, Mobile and WoT: Web performance, measurement, and characterizationSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Markerless motion capture enables the tracking of human motion without requiring physical markers or suits, offering increased flexibility and reduced costs compared to traditional systems. However, these advantages often come at the expense of higher computational demands and slower inference, limiting their applicability in real-time scenarios. In this technical report, we present a fast and reliable neural inverse kinematics framework designed for real-time capture of human body motions from 3D keypoints. We describe the network architecture, training methodology, and inference procedure in detail. Our framework is evaluated both qualitatively and quantitatively, and we support key design decisions through ablation studies.

Problem

• Real-time human motion capture from 3D keypoints
• Reducing computational demands of markerless systems
• Improving speed and reliability of inverse kinematics

Innovation

• Fast neural inverse kinematics framework
• Real-time 3D keypoint motion capture
• Detailed network architecture and training