Abstract: Based on optical motion capture, two joint center estimation models: Bi-segment
Model and Sphere Model are given by analyzing the motion of segments constrained by joints.
For each model a least squares solution is presented respectively and the joint center location
could be acquired by solving a linear equation. A virtual linkage is used to test the accuracy of the
two least squares solutions. Using the Bi-segment Model which is more accurate, a technique for
virtual human calibration is proposed based on the topology of the virtual human. The comparison
of measured values and calibrated values of a performer demonstrates errors for limbs are less
than 4.3%. Finally, the calibrated virtual human is driven. The result shows the virtual human can
mimic the performer vividly and limbs intersection cases are reduced.

Key words: optical motion capture, joint center estimation, least squares solution, virtual
human calibration