Human Skeleton Modeling from 2D Uncalibrated Monocular Data

Abstract

A novel method which is simple but effective is proposed to estimate human skeleton ratios from 2D uncalibrated monocular data. Unlike the existing skeleton estimation methods where pre-defined models are used or identification of body segments with certain attributes is necessary, the proposed method utilizes only the 2D joint locations as the input source without posture estimation. In addition, the proposed method uses a real perspective camera model instead of the popularly-used scaled orthographic camera model. The proposed method is tested on monocular data from different camera motions and the estimation result is satisfactory. The reconstructed human skeleton model can be used in further research for full body reconstruction or motion reconstruction from monocular data.