Camera calibration from human motion

This paper presents a method for the self-calibration of non-rigid affine structure to a Euclidean co-ordinate frame from only two views by enforcing constraints derived from the known structure of the human body, such as piecewise rigidity and approximate symmetry. We show that the proposed algorithm is considerably more efficient yet equally accurate when compared to previous methods. The resulting structure and motion is then refined further using a full bundle adjustment to give maximum likelihood values for body segment lengths and joint angles. A quantitative analysis is presented using synthetic data whilst qualitative results are demonstrated for real examples of human motion.