RELIABILITY AND VALIDITY OF A DEEP LEARNING ALGORITHM BASED MARKERLESS MOTION CAPTURE SYSTEM IN MEASURING SQUATS

Authors

Sakurako Musha, Sportip, Inc., Tokyo, JapanFollow
Daishiro Kobayashi, Sportip, Inc., Tokyo, JapanFollow
Yuya Takaku, Sportip, Inc., Tokyo, JapanFollow
Yasuko Hirono, Faculty of Health and Sport Sciences, University of Tsukuba, JapanFollow
Takuya Otsu, Faculty of Health and Sport Sciences, University of Tsukuba, JapanFollow
Norihisa Fujii, Faculty of Health and Sport Sciences, University of Tsukuba, JapanFollow

Category

Technology/equipment

Document Type

Paper

Abstract

This study aimed to compare the performance of a traditional marker-based motion capture system and a video-based markerless system in analyzing squats and to determine the reliability and validity of the markerless system. Twenty-one squats were recorded using a marker-based motion capture system and a 2D video camera. We analyzed the 2D video data using Sportip Motion 3D, a deep learning-based 3D human pose estimation algorithm based specifically on sports activities, and the peak lower limb joint angles were calculated by both systems. There was an excellent agreement between VICON and Sportip Motion 3D for all joint angles (hip intraclass correlation coefficient (ICC) = 0.96, knee ICC = 0.92, ankle ICC = 0.86), with average differences of less than 1.3°. These results indicate that squat analysis using Sportip Motion 3D is equally reliable and accurate as the conventional marker-based method.

Recommended Citation

Musha, Sakurako; Kobayashi, Daishiro; Takaku, Yuya; Hirono, Yasuko; Otsu, Takuya; and Fujii, Norihisa (2022) "RELIABILITY AND VALIDITY OF A DEEP LEARNING ALGORITHM BASED MARKERLESS MOTION CAPTURE SYSTEM IN MEASURING SQUATS," ISBS Proceedings Archive: Vol. 40: Iss. 1, Article 122.
Available at: https://commons.nmu.edu/isbs/vol40/iss1/122