A Survey of Recent Advances on Two-Step 3D Human Pose Estimation

Abstract

Human pose estimation in images is an important and challenging problem in Computer Vision. Currently, methods that employ deep learning techniques excel in the task of 2D human pose estimation. 2D human poses can be used in a diverse and broad set of applications, of great relevance to society. However, the use of 3D poses can lead to even more accurate and robust results. Since joint coordinates for 3D poses are difficult to estimate, fully convolutional methods tend to perform poorly. One possible solution is to estimate 3D poses based on 2D poses, which offer improved performance by delegating the exploration of image features to more mature 2D pose estimation techniques. The goal of this paper is to present a survey of recent advances on two-step techniques for 3D human pose estimation based on 2D human poses.