en-de
en-es
en-fr
en-sl
en
en-zh
0.25
0.5
0.75
1.25
1.5
1.75
2
Learning to Reconstruct 3D Human Pose and Motion from Silhouettes
Published on Feb 25, 20075881 Views
We will describe our ongoing work on learning-based methods for recovering 3D human body pose and motion from single images and from monocular image sequences. The methods work directly with raw image
Related categories
Chapter list
Learning to Reconstruct 3D Human Pose and Motion from Silhouettes 00:01
Goal00:17
2 Broad Classes of Approaches01:00
“Model Free” Learning – based Approach02:54
The Basic Idea04:02
Silhouette Descriptors04:56
Why Use Silhouettes ?05:03
Ambiguities06:19
Shape Context Histograms07:23
Shape Context Histograms Encode Locality10:18
Nonlinear Regression11:22
Regression Model11:31
Regularized Least Squares12:59
Relevance Vector Machine … a brief introduction14:00
Contd.16:19
Pose from Static Images17:25
Training & Test Data17:30
Methods Tested18:59
Synthetic Spiral Walk Test Sequence19:49
Spiral Walk Test Sequence20:52
Some statistics .. 21:58
Glitches22:56
TITLE23:45
Real Image example24:06
Understanding the Problem24:19
Pose from Video Sequences25:00
Tracking Framework25:03
Joint Regression equations25:35
Results with Joint Regression26:32
Spiral Walk Test Sequence26:45
Real Images Test Sequence27:25
Conclusion27:41
Real Images Test Sequence30:43