Tracking articulated motion of a moving human via markered motion capture involves a model of the skeleton and how it moves. Conventionally the movement is parameterised by Euler angles and the tracking is performed by an extended Kalman filter. If instead we choose to track the bivector components representing the rotational movements of the limbs, we require both the forward kinematics and derivatives with respect to the bivectors as input to the Kalman filter. We present methods of producing such inputs and compare the performance of the two parameterisations.

jl@eng.cam.ac.uk