Learning Basis Representation to Refine 3D Human Pose Estimations

Published in AAAI, 2019

Estimating 3D human poses from 2D joint positions is an ill-posed problem, and is further complicated by the fact that the estimated 2D joints usually have errors to which most of the 3D pose estimators are sensitive. In this work, we present an approach to refine inaccurate 3D pose estimations. The core idea of the approach is to learn a number of bases to obtain tight approximations of the low-dimensional pose manifold where a 3D pose is represented by a convex combination of the bases. The representation requires that globally the refined poses are close to the pose manifold thus avoiding generating illegitimate poses. Second, the designed bases also have the property to guarantee that the distances among the body joints of a pose are within reasonable ranges. Experiments on benchmark datasets show that our approach obtains more legitimate poses over the baselines. In particular, the limb lengths are closer to the ground truth.


Robust 3D Human Pose Estimation from Single Images or Video Sequences

Published in T-PAMI, 2018

This paper estimate 3D human poses from a monocular image or video. It learns the anthropometric priors of human poses by sparse coding and optimizes simultaneously the camera parameters and 3D poses. It outputs multiple 3D pose estimation for a single subject and proposes to use the temporal information to select the optimal one.


Minining 3D Key-Pose-Motifs for Action Recognition

Published in CVPR, 2016

This paper mines a set of representative short sequences for each action. A pose sequence is classified by matching the sequence to the motifs of each class. The class achieves the least error is the predicted class. This simple method achieves the state-of-the-art performance on several benchmark dataset.