Host: The Japan Society of Mechanical Engineers
Name : [in Japanese]
Date : May 27, 2020 - May 30, 2020
This paper extracted features that do not depend on posture from body shape, and worked on personal authentication using the extracted features. Heat Kernel Signature is a method of extracting features that do not depend on posture from the body shape. Using the shape descriptor that is resistant to non-rigid deformation, the same part between humans was searched, and the part was identified as a feature. The proposed method achieved 96.9% accuracy when narrowing down from 20 to 3 people. We also verified this method for actual walking. By transforming the previously acquired skin model so that it fits the three-dimensional joint position acquired by Video Motion Capture, the body shape can be reconstructed from the footage image. As a result, the accuracy of identifying one person out of four was 88.75%
