A Comparative Study of Recent 2D Human Pose Estimation Methods

Abstract

Human Pose Estimation is a computer vision technique utilized in various fields such as healthcare, security, and sports to detect the pose of single or multi-person utilizing various machine learning and deep learning methodologies. These methodologies can be applied in a variety of ways, including top-down and bottom-up approaches. Furthermore, this technique can be used with images or videos, as well as in 2D or 3D domains. This paper delves into recent advancements in deep learning methods used for estimating the pose, covering commonly employed datasets, evaluation metrics, and the challenges encountered in this field. It aims to assist new researchers in getting familiar with this technique based on work published between 2019 and 2023, with a spotlight on 2D and omitting 3D.