This paper describes a method for estimating three dimensional (3D)positions and orientations of vertebrae together with the load working on the vertebral bodies from a sequence of human images taken by a video camera based on a 3D whole body mode and a 3D spine model. In this method, we first obtain the neck and waist postures, the centers of masses of the head, trunk, and arms by fitting the 3D whole body model to an input human image. Then the 3D spine model is deformed so that it satisfies equilibrium of the force working on the vertebrae caused by the springs which approximate intervertebral discs. After determining the spine posture, the load working on the top surfaces of the vertebral bodies is calculated based on the centers of masses and the weights of the body parts. The experimental results based on one real MR image dataset showed that our method estimated the vertebra positions within the positional shifts of 6.2mm and the rotational variation of 3.1 degrees. We also confirmed that our method estimated the spine posture and calculated the force working on the top surface of the lumbar vertebra L, 5' from the real human video images reasonably.
