Spine posture estimation method from human images using 3D spine model computation of the rough approximation of the physical forces working on vertebral bodies

This paper describes a method for estimating a human spine posture from human images using a human spine model that is possible to compute the rough approximation of the physical forces working on vertebral bodies. The spine posture estimation model is composed of the vertebral bodies, each of which is modeled as a rigid body, and the intervertebral discs that are modeled as springs. Our method uses the positions of the neck and waist in addition to the positions of the head, torso, and arms estimated from the actual human images. The spine model is deformed so as to locate the top and the bottom vertebrae of the spine model to the estimated neck and waist positions. According to the experiments based on one real MR image dataset of one subject person, our methods estimated the positions of the vertebrae within positional shifts of about 6.3 mm and the rotational variation of about 3.1 degrees. We also confirmed the methods calculated the reasonable estimation of the physical forces working on the vertebral body.