Compensating for patient motion constitutes a challenging as well as important problem for high precision cranial radiotherapy. Common approaches rely on patient immobilization using stereotactic frames, masks or bite bars, or employ X-ray based monitoring and motion compensation such as the 6D skull tracking used by the CyberKnife®. These techniques entail several drawbacks: First, the mask systems are rather inconvenient and also not tolerated by all patients. They further provide no motion monitoring during the treatment - the patient is assumed to be fixed. This is indeed only true up to errors in millimeter range. Second, X-ray based imaging exposes the subject to an unnecessary amount of additional radiation. This exposure also limits the head-tracking speed to about 1 Hz.

Therefore, marker-less optical head-tracking provides a promising alternative, where a laser constantly scans the patient's forehead. This approach requires only light patient fixation  and  provides a basis for fast real-time monitoring. While promising results have already been achieved using commercial systems such as the Microsoft Kinect®, the accuracy still lags behind the desirable standard of sub-millimeter accuracy. One major drawback of marker-less tracking is given by the lack of point-to-point correspondences. Finding these, for objects which are not strictly rigid, is challenging and prone to registration errors. Point cloud matching algorithms are used which rely, if at all, on a few spatial landmarks. The surface geometry very often introduces further ambiguities which lead to a convergence into local error minima and hence poor robustness.