Motion capture systems using infrared cameras (Marker-Based MoCap) are most frequently used in biomechanics research because they can accurately measure the 3D coordinates of reflective markers. However, if reflective markers cannot be affixed to the body or data cannot be collected from a large number of subjects in a short period of time, Marker-Based MoCap cannot be used; thus, it is necessary to collect movement data without using markers. The purpose of this study was to develop a markerless motion capture method using a human body model and a neural network for silhouette extraction, which are open source. We aimed to compare its accuracy with that of previously developed markerless motion capture methods, and to examine the superiority of the proposed method. Thirteen subjects walked at arbitrary speeds, and their movements were captured by eight RGB cameras. Based on the obtained images, the 3D coordinates of the joint centers were calculated using the proposed method of matching the silhouettes with the human body model. The mean error of the joint center was less than 30 mm for all joints, except for the hip joint, which had an error of 38.7 mm. Comparing the proposed method to markerless motion capture using a 3D scanner, the proposed method was less accurate for most joints. Therefore, in experimental environments where a 3D scanner is available the proposed method should not be used. The accuracy of the proposed method was higher than that of the commercial markerless motion capture (Theia 3D), except for the hip and hand joints. Therefore, the investigator must rationally select between the two methods according to the purpose of the research and the experimental environment. In comparison with markerless motion capture using an open source (e.g. OpenPose), the proposed method was more accurate, except for the hip joint; therefore, it would be better to choose the proposed method in most cases. The accuracy of the proposed method was particularly low at the center of the hip joint. It would be necessary to improve the hip joint accuracy by identifying the superior anterior iliac spine, superior posterior iliac spine, and greater trochanter in the human body model.
© Copyright 2025 Japan Journal of Physical Education, Health and Sport Sciences. Japan Society of Physical Education, Health and Sport Sciences. All rights reserved.