A Movement Analysis Application using Human Pose Estimation and Action Correction

Abstract

Human pose estimation (HPE) is an important field of computer vision that aims to predict poses of individuals from videos and images. It has been used in many different areas including human-computer interaction, motion analysis, surveillance, action prediction, action correction, augmented reality, virtual reality, and healthcare. Executing movements correctly is crucial in all kinds of physical activities, both to increase performance and reduce risk of injury. HPE is poised to help athletes better analyse the quality of their movements. This work proposes a model for movement analysis, repetition count, and movement correction in physical exercises using HPE. For this purpose, a study is carried out in the field of HPE applied to sports and another study is focused on HPE for correction and postural analysis. From this, it is verified what is the state of the art in HPE for physical exercises and what is the best method for analyzing movements. This work implements an application with improvements in respect to other related work, focusing mainly on the feedback presented to the user when performing a certain movement. To validate the proposed model, a quantitative research was carried out using the Unified Theory of Acceptance and Use (UTAUT). For both people who exercise and professionals in the field of physical education, the results demonstrate that the application is able to analyze the biomechanics of movement, responding with speed and precision to execution errors. Among other results are: user satisfaction, interest in using the application in the future, and agreement in relation to good performance in helping and analyzing physical exercises.