Calibration Accuracy Evaluation for Stereo Reconstruction

Abstract

During SPECT imaging, a patient must remain motionless for 20 minutes or more. Without compensation, patient motion can compromise diagnostic accuracy. As the first step in compensation, we propose stereo-optical tracking of retro-reflective spheres attached to stretchy bands wrapped about the patients to estimate patient motion. Camera calibration, which is a crucial step in stereo-imaging, ultimately determines the accuracy of tracking. The two leading methods of camera calibration are: 1) phantom-based calibration, and 2) grid board-based calibration. Both use the Tsai camera model, but use different input data and methods of calculation. Thus, the two methods show different sensitivities to system modeling and input data inaccuracies. We evaluated the relationship between inaccuracies using two error measures. The first was the residual error or the difference between the input center location of spheres in the 2D optical images and the centers estimated using the calibration parameters. The second was stereo error or the difference between the input 3D location of the center of spheres in the calibration phantom and the centers estimated by stereo imaging. We observed that grid board-based calibration is more stable, and that for both methods the stereo error increases with distance away from the center of the measurement volume.