Automatic Calibration of Magnetic Motion Trackers using a Bayesian-Neural Cascade

Motion capture data are being used to address an increasingly broadening repertoire of animation related tasks, ranging from film production and game products to research on realistic human motion and even medical applications. Because of their vast utility, many technologies have been developed to acquire such data but magnetic trackers remain the dominant motion capturing solution for computer graphics applications. But even though magnetic trackers have many advantages versus their mechanical, optical and sonic counterparts, they are amenable to electromagnetic interference. Conventional calibration procedures are tedious and very time consuming, since they require the manual collection of measurements on a relatively dense three dimensional grid inside the tracking area. Despite this labor-intensive procedure, this type of calibration cannot compensate for every type of distortion resulting in limited accuracy and even motion discontinuities in noise contaminated environments. We propose a Bayesian-neural approach that allows the automatic calibration of magnetic motion capturing systems, using motion tracking data from a short, free-form tracking of a calibration pole.