Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : June 02, 2018 - June 05, 2018
In this paper, we propose the inverse kinematics using the high-order information moments and build video motion capture system with multiple cameras. The high-order information moments are what combines the high-order moments from Part Confidence Maps showing the existing probability of joints in an image with ones from a human silhouette image. In this system, there're two steps. First one is rough estimation and the next one is precise estimation. Rough estimation is estimating 3d position of major human joints. Precise estimation is estimating 3d position of other parts including hands and feet. For rough estimation, we achieved 6.76cm about the average joint error over 740 jumping motion. For precise estimation, by dividing a whole body into some parts, we built the technical basis about video motion capture considering a whole body including hands and feet.
