CT auto-calibration by consistent contours

This paper discusses a novel strategy for estimating projection matrices of a computed tomography using object contours in projection images. We formulate mathematical conditions for projection parameter consistency based on contour detection and tracking in 2D. We derive a global optimization problem and a greedy optimizer herein. Applying the approach for motion detection shows an average reduction of motion error in object space down to 0.15mm in a standard dental cone-beam CT setup. Evaluation is performed on simulated data with different motion types and object shapes, and considers the influence of noisy contour data. The accuracy of estimation is linked to sample resolution, number and distinctiveness of imaged objects, and accuracy of input data. Our strategy is robust towards inaccurate initialization and truncated data.