Bayesian Image Based 3D Pose Estimation
We introduce a 3D human pose estimation method from single image, based on a hierarchical Bayesian non-parametric model. The proposed model relies on a representation of the idiosyncratic motion of human body parts, which is captured by a subdivision of t
- PDF / 2,888,488 Bytes
- 17 Pages / 439.37 x 666.142 pts Page_size
- 86 Downloads / 257 Views
Abstract. We introduce a 3D human pose estimation method from single image, based on a hierarchical Bayesian non-parametric model. The proposed model relies on a representation of the idiosyncratic motion of human body parts, which is captured by a subdivision of the human skeleton joints into groups. A dictionary of motion snapshots for each group is generated. The hierarchy ensures to integrate the visual features within the pose dictionary. Given a query image, the learned dictionary is used to estimate the likelihood of the group pose based on its visual features. The full-body pose is reconstructed taking into account the consistency of the connected group poses. The results show that the proposed approach is able to accurately reconstruct the 3D pose of previously unseen subjects.
Keywords: Human pose estimation Bayes
1
·
Hierarchical non-parametric
Introduction
Human pose estimation from images has been considered since the early days of computer vision and many approaches have been proposed to face this quite challenging problem. A large part of the literature has concentrated on identifying a 2D description of the pose mainly by trying to estimate the positions of the human joints in the images. Recently, attention has been shifted to the problem of recovering the full 3D pose of a subject either from a single frame or from a video sequence. Despite this is an ill-posed problem due to the ambiguities emerging by the projection operation, the constraints induced by both human motion kinematics and dynamics have facilitated the recovery of some accurate 3D human pose estimation. In this work we approach the problem of 3D pose estimation from a single image building a hierarchical framework based on Bayesian non-parametric estimation. A schema of the framework is shown in (Fig. 3). Following the schema flow, we divide the human body into different parts and we study the idiosyncratic motion behavior of each part independently from the others. In this way Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46484-8 34) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part VIII, LNCS 9912, pp. 566–582, 2016. DOI: 10.1007/978-3-319-46484-8 34
Bayesian Image Based 3D Pose Estimation
567
G1 Pose Dictionary Visual Dictionary
G12 Pose Dictionary Visual Dictionary
Query image
Detected 2D joints
Dictionary based group 3D pose estimation
Final result
Fig. 1. Method overview; 3D pose estimation given a query image.
we learn the principal motion modes of each part. Each body part is specified by a group of joints, and its motion is represented by pose features obtained by the principal motion direction on the SE(3) manifold with respect to a reference pose. As a natural reference pose we consider the “Vitruvian man” pose presented in Fig. 2 together with the selected groups. The visual features for each group are the PHOG features of [1], which are computed using t
Data Loading...
