Table 1 Comparative analysis between traditional methods and proposed system.
Dimension | Traditional methods | P-STMO[25] | MotionBERT[27] | HiEve[28] | Our proposed system |
|---|---|---|---|---|---|
Real-time processing | Immediate but subjective | 8–10 fps | 12–15 fps | 15–18 fps | 35.4 fps |
Multi-modal feedback | Verbal/visual only | None | None | None | AR + audio + haptic |
Evaluation objectivity | Subjective coaching[12] | 42.1Â mm MPJPE | 93.0% accuracy | Multi-task | 92.4% on cheerleading |
Personalization | Experience-based[12] | Generic | Generic | Generic | Bayesian skill adaptation |
Scalability | 1:1 coaching ratio[13] | Unlimited | Unlimited | Unlimited | Unlimited with sport-specific |
Multi-person analysis | Manual observation[14] | Limited | Good | Excellent | Transformer-based (85.3% PCK) |
Learning efficiency | Baseline (100%)[15] | N/A | N/A | N/A | 35% faster acquisition |
Occlusion handling | Visual judgment[14] | 62.7% PCK | 68.3% PCK | 75.2% PCK | 85.3% PCK |
Cognitive load Mgmt | Intuitive[16] | Not addressed | Not addressed | Not addressed | NASA-TLX optimized |
Deployment platform | In-person only[13] | Desktop/GPU cluster | GPU cluster | Multi-camera setup | Web/mobile/desktop |
