Accuracy Verification of 3D Motion Analysis System Using Smart-phone Monocular Camera

Abstract

PURPOSE This study aimed to verify the accuracy of three-dimensional (3D) motion data produced through artificial intelligence-based user motion recognition technology with images obtained using a smartphone monocular camera. This was done to explore the possibility of developing an application that can improve the reliability of the measurement of physical activity performing motions and feedback provision. METHODS To check the accuracy of the artificial intelligence-based 3D motion analysis system that utilized a semi-supervised learning method, a commercialized 3D infrared motion analysis system measured and compared motions on three movement planes, motions with limited joint movement, and fast motions in a wide moving range. RESULTS The motions on the coronal and sagittal planes produced through the 3D motion analysis application showed very high measurement accuracy; however, the accuracy of the measurement of motions on the horizontal plane, which could not be measured directly with a camera, was relatively lower than that of the coronal and sagittal planes. Accuracy in measuring 3D motion was moderate in moving motions and low in motions with limited joint movement. CONCLUSIONS For the developed 3D motion analysis system to be used in online physical education, the types of physical activities included in the program should be comprehensively composed through the analysis of the content system of the physical education curriculum and the resultant physical activities.

Keywords

• online physical education
• 3d motion analysis
• smart phone application
• semi-supervised learning