Visual marker localization in robot-assisted stereotactic neurosurgery

An approach is developed to automatically detect and locate the visual markers glued on the head in robot assisted stereotactic neurosurgery. This approach has four major steps. At first, one special marker, defined as “B” marker, is searched in the entire image. By this the search range can be decreased because the distance of markers between each other cannot exceed the patient's head. Secondly, in the local image the candidate markers are searched which are marker-like connected region. Their features are extracted and the false markers are discarded. Thirdly, the method will try to find the missing markers with epipolar constraint in binocular cameras if some markers are lost. Finally the markers' can be located in space with stereopsis. We use four markers that are glued to the skin of head in our experiment. The markers' features are extracted using 82 pairs of images and the approach is tested on 166 pairs of other images. The false marker rate is only 0.08%. The approach can be used in head motion tracking and automatic image registration in robot-assisted stereotactic neurosurgery.