Motion is a prominent artefact in intraoperative neurosurgical imaging. We are investigating feature-based algorithms to detect and correct motion in visual-light images. This paper presents a new multilevel interpolation method to reconstruct motion of the brain surface from the displacement of image features. The proposed method proved to be able to sufficiently correct all motion artefacts. Compared to other recently derived methods it is more stable against outliers and more reliable for sparse data.