Document Type



This study presents a method for estimating human 3D body shape in action. We propose a method for estimating 3D human body shape motion that uses multiple view images and visual hulls. Related methods necessitated lengthier preparations, such as camera calibration, which would require several tries before actually capturing the image. We solve this issue by combining state-of-the-art computer vision methods to automatically process the required inputs and parameters, so that camera images are the only resource needed for estimation. In our experiments, we applied our method to a video of human subject kicking a soccer ball to left and right side of a goal; we successfully acquired the subjects 3D body shape. In addition, we verified that the application’s automatically obtained body shape successfully provides the subject’s center of mass.

New Investigator Award