Bioinformatics and Medical Applications

8An Insight Into Human Pose Estimation and Its Applications

Shambhavi Mishra¹*, Janamejaya Channegowda² and Kasina Jyothi Swaroop³

¹CBP Government Engineering College, Delhi, India

²Ramaiah Institute of Technology, Bengaluru, India

³Indian Institute of Technology (Indian School of Mines), Dhanbad, India

Abstract

Human pose estimation has been an active research area in computer vision and has received great attention till date. Motion tracking and activity recognition are some of the applications that utilize pose estimation. We define human pose estimation as localization of human joints (these joints are called key points—elbows, wrists, etc.) as well as labeling them in our data comprising either images or videos. In this chapter, we will describe fundamentals of human pose estimation and study different research approaches available in literature.

We will discuss different categories within pose estimation and mention key differences between them. The chapter will begin by reporting classical methods of pose estimation such as human pose recognition using motion segmentation and as well as deep learning–based methods explored in recent years. The chapter will also touch upon drawbacks of classical models and the evolution of Convolutional Neural Networks developed to overcome these shortcomings.

Keywords: Action recognition, pose estimation, pose detection, human activity recognition, human pose estimation, 2D pose estimation, 3D pose estimation, body joints localization ...