The problem of human pose estimation, defined as the problem of localization of human joints. Human pose estimation in this work provide the method that not using Deep Learning approach as much as possible.
- Background Subtraction, Morphology, Canny Edge Detection, Convex Hull
- Reduce time used for the classification
- X
- Y
- Alpha
From the implementation has showed that trained SVM classifier will have a good accuracy
only when the image have a clear depth estimation value.
In this work estimate 16 key points: Head, Neck, Spine, Pelvis, r_Ankle, l_Ankle, r_Knee, l_Knee, r_Hip, l_Hip, r_Shoulder,
l_Shoulder, r_Elbow, l_Elbow, r_Wrist, l_Wrist
Head | l_Shoulder | r_Shoulder | l_Elbow | r_Elbow | l_Wrist | r_Wrist | Neck | |
---|---|---|---|---|---|---|---|---|
All | 10 | 10 | 10 | 10 | 10 | 8 | 9 | 9 |
Correct | 7 | 4 | 2 | 2 | 0 | 1 | 1 | 4 |
% | 70.00% | 40.00% | 20.00% | 20.00% | 0.00% | 12.50% | 11.11% | 44.44% |
Spine | Pelvis | l_Hip | r_Hip | l_Knee | r_Knee | l_Ankle | r_Ankle | |
---|---|---|---|---|---|---|---|---|
All | 9 | 8 | 10 | 10 | 10 | 10 | 9 | 8 |
Correct | 8 | 3 | 5 | 4 | 0 | 0 | 2 | 2 |
% | 88.88% | 38.00% | 50.00% | 40.00% | 0.00% | 0.00% | 22.22% | 25.00% |