Agentic Pipeline for Self-Synchronized Multiview Joint Angle Monitoring
A team of researchers has introduced an innovative pipeline for monitoring joint angles from multiple views in uncalibrated settings, utilizing two cameras that operate without hardware triggers. This system integrates multimodal large language models for seamless video synchronization and self-verification driven by agents. Advanced monocular 2D pose estimation techniques are employed to generate candidate poses, while an agent-based selection process effectively identifies and tracks the target individual, ensuring reliable 2D pose outputs even in the presence of multiple subjects. This method tackles the difficulties associated with kinematic monitoring in spinal cord injury patients, as markerless motion capture techniques show promise but struggle with synchronization and calibration challenges.
Key facts
- arXiv:2605.16419v1
- Two cameras without hardware triggers
- Multimodal large language models enable automatic video synchronization
- Agent-driven self-verification
- State-of-the-art monocular 2D pose estimation models
- Agent-based selection mechanism for target subject tracking
- Addresses calibration and synchronization challenges
- For spinal cord injury rehabilitation
Entities
Institutions
- arXiv