A survey of image registration techniques
ACM Computing Surveys (CSUR)
IEEE Transactions on Pattern Analysis and Machine Intelligence
Multi-stage combination of geometric and colorimetric detectors for eyes localization
ICIAP'05 Proceedings of the 13th international conference on Image Analysis and Processing
| Hi-index | 0.00 |
We propose an image-based visual servoing of a 6 dof manipulator robot used for patient alignment in protontherapy. In the case of intra-skull tumour treatment, patient's head is fixed to a couch by a plastic mould contention mask. The couch is carried and displaced by the end-effector of the robot, in order to position the tumour with respect to the proton beam isocenter. The end-effector velocity is computed on the basis of four image feature points. In a pre-operatory calibration, patient's head is correctly positioned, which is ensured by use of two X-rays shot of the skull, and a reference image of the head is stored as a template of correct alignment. The image is divided into four quadrants: the center of each quarter defines a goal feature, this set of points defines a desired feature vector of eight elements. Each quadrant is modeled by a combination of its grey levels, edges and a set of corners. The feature points are further matched using a template matching strategy with a window sliding around the reference location. In a daily patient repositioning, the translation between each quadrant center w.r.t its template is computed to build the image error function. A proportional control law is used to compute the manipulator velocity w.r.t to the error function and the image Jacobian. The complete algorithm runs at 200 ms/frame on a standard PIV PC. The aim of that video-based patient repositioning could avoid X-ray verification of patient alignment, reducing patient dose and duration of treatment sessions.