Glyph-Mocap Fusion: hybrid framework for preschool dance synthesis

The Glyph-Mocap Fusion framework integrates multiscale glyph feature extraction with optical motion capture to address challenges in preschool dance choreography. Using infrared cameras and RGB-D sensors for high-precision data acquisition, the system enhances biomechanical accuracy and semantic fidelity. A CNN-Transformer-based StrokeNet module processes glyph hierarchies, while a kinematics-conditioned Motion Variational Autoencoder (MotionVAE) generates anatomically accurate and personalized movements. Optical imaging further refines motion generation by leveraging stereo vision and depth data to boost semantic alignment. Evaluations with Fréchet Inception Distance (FID), Mean Per Joint Position Error (MPJPE), and cross-modal alignment accuracy show significant improvements over traditional 3D motion capture methods. The system achieves real-time performance on NVIDIA Jetson TX2, ensuring low resource utilization while enabling augmented reality (AR)-based visualization to enhance the learning experience. This work provides a scalable, technically robust, and culturally enriched solution for preschool education, advancing the creation of engaging and precise choreography for young learners.