DMCA
3D Human Pose from Silhouettes by Relevance Vector Regression (2004)
Cached
Download Links
| Venue: | In CVPR |
| Citations: | 199 - 8 self |
BibTeX
@INPROCEEDINGS{Agarwal043dhuman,
author = {Ankur Agarwal and Bill Triggs},
title = {3D Human Pose from Silhouettes by Relevance Vector Regression},
booktitle = {In CVPR},
year = {2004},
pages = {882--888}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
We describe a learning based method for recovering 3D human body pose from single images and monocular image sequences. Our approach requires neither an explicit body model nor prior labelling of body parts in the image. Instead, it recovers pose by direct nonlinear regression against shape descriptor vectors extracted automatically from image silhouettes. For robustness against local silhouette segmentation errors, silhouette shape is encoded by histogramof-shape-contexts descriptors. For the main regression, we evaluate both regularized least squares and Relevance Vector Machine (RVM) regressors over both linear and kernel bases. The RVM’s provide much sparser regressors without compromising performance, and kernel bases give a small but worthwhile improvement in performance. For realism and good generalization with respect to viewpoints, we train the regressors on images resynthesized from real human motion capture data, and test it both quantitatively on similar independent test data, and qualitatively on a real image sequence. Mean angular errors of 6–7 degrees are obtained — a factor of 3 better than the current state of the art for the much simpler upper body problem. 1.
Keyphrases
human pose relevance vector regression kernel base local silhouette segmentation error main regression silhouette shape explicit body model worthwhile improvement single image least square shape descriptor vector body problem similar independent test data human body real human motion capture data real image sequence prior labelling histogramof-shape-contexts descriptor mean angular error direct nonlinear regression relevance vector machine monocular image sequence image silhouette much simpler current state body part good generalization
