Pose Estimation of Multiple X-ray Cameras Using Planar Markers and Epipolar Consistency

This paper introduces an image-based approach for estimating pose of multiple X-ray cameras capturing a diagnostic part such as a wrist, in an orthopedic surgery. The proposed method uses a calibration plane in which five fiducial lead-markers are positioned in a parallelogram pattern. We assume that the calibration plane is attached to the target part and observed in every view. Initial camera extrinsic parameters are calculated from planar homography, followed by a nonlinear optimization using reprojection error. Views with high reprojection error are further refined using epipolar consistency. We evaluate our method by simulating a Conebeam CT trajectory using digitally reconstructed radiographs from a CT model of a wrist phantom.