Introduction
Related works
Multi-view 3D pose estimation
Sensor-based jump analysis in figure skating
Manual-observation-based jump analysis in figure skating
Framework of the reconstruction system
Likelihood distribution and temporal smoothness-based discrete probability points selection likelihood
Likelihood distribution
Temporal smoothness
Multi-perspective and combination unification-based large-scale venue 3D reconstruction
Multi-perspective
Camera combinations unification
Multi-constraint-based human skeleton estimation
Priors based on body structure and motion trend
Multi-constraint
Experiment result
Data set and experimental environment
Verification test of camera number
Evaluation method
Experimental results
Items | 2D | Reconstruction | 3D |
---|---|---|---|
Basic framework | Joint pixel value | Reconstruct with stable camera combinations + calculate average value | Without any correction |
Conventional work [21] | Joint pixel value + temporal smoothness | Reconstruct with stable camera combinations + calculate average value | With simple prior-based correction |
Proposed method | Joint confidence map + temporal smoothness | Reconstruct with all camera combinations + calculate the unified value of combinations | With multi-constraint-based correction |
Items | Basic framework | Conventional work [21] | Proposed method | |
---|---|---|---|---|
Error range @30 mm | Upper | 40.97% | 70.60% | 88.89% |
Lower | 38.73% | 80.4%8 | 95.08% | |
Total | 40.01% | 74.83% | 91.55% | |
Error range @50 mm | Upper | 47.98% | 72.97% | 90.26% |
Lower | 41.02% | 84.23% | 95.65% | |
Total | 44.99% | 77.77% | 92.57% | |
Error range @70 mm | Upper | 51.75% | 75.82% | 91.30% |
Lower | 43.04% | 86.68% | 96.15% | |
Total | 48.02% | 80.48% | 93.38% | |
MPJPE (mm) | Upper | 112.11 | 85.47 | 28.74 |
Lower | 124.59 | 60.75 | 21.56 | |
Total | 121.45 | 74.12 | 23.57 |