Video-Based Human Action Recognition and Its Application in Dance Research

In this chapter, Human Action Recognition (HAR) is explained as a method for motion detection in the context of machine learning and computer vision methods. The main goal of this method is to investigate the recognition accuracy of single person actions in videos. The paper focuses on the clear description of the underlying methods such as “Optical Flow”, “Pose Estimation” and “Skeleton Data”, which were applied in an interdisciplinary research project for the determination of positions and motion features in dance (including dance motion sequences).