Human spine posture estimation from video images based on connected vertebra spheres model

This paper reports a method for estimating human spine posture using the front and side views of a human body taken by a video camera. We present here a new 3D model to estimate the spine posture using connected vertebra spheres. In this model, each vertebra forming the spine is approximated as a sphere. The spine is approximated as a series of spheres connecting each other. Each sphere has the control points to represent the outer shape of the body. The spine posture is estimated by moving the sphere and the control points. The estimation is performed by calculating the matching ratio between a projected image of the model and an input image. X-ray CT slices are used for the construction of the model. We applied the proposed method to real human images. The experimental results show that our 3D model worked reasonably well for the estimation of human spine posture based on real human images.