Augmented reality, hand–eye calibration, patient–image registration, endoscope, optical tracking system
I proposed a one-step calibration method to simultaneously perform patient-to-image registration and hand-eye calibration that are necessary processes for implementing an augmented reality (AR) navigation using an optical tracking system in endoscopic surgeries. Until now, the registration and calibration processes have been separately performed, called the two-step calibration method here. The two-step method requires the different condition from the AR one for performing it. Although the optimal solutions from the two-step method are calculated under their conditions, they are not optimal for the AR condition due to multiplication errors as well as tracking errors. To alleviate the errors, a transformation from the coordinate system of the endoscope to that of the computed tomography images is accurately calculated and used in the one-step method. Because this transformation is free from the multiplication and tracking errors, it becomes an important factor to improve the AR accuracy. A series of experiments with synthetic and real data were performed. Considering the real characteristics of system errors, the synthetic data was generated. The effects of tracking noise, feature noise, the number of fiducial markers, and the number of poses were analyzed through the simulation. For experiments with real data, we used a phantom designed considering real sinus and skull-base surgeries, and the AR error of the one-step method was compared with that of the two-step method. The comparison results showed the significantly increased accuracy of the one-step method.
Table Of Contents
I. INTRODUCTION 1 1.1 Introduction to Augmented Reality 1 1.2 AR based Surgical Navigation 3 1.3 Challenges and Related Works 8 1.4 Thesis Contributions 9 Ⅱ. AR IMPLEMENTATIONS 10 2.1 Nomenclatures 11 2.2 AR Configuration 12 2.3 Camera Calibration 14 Ⅲ. TWO-STEP CALIBRATION METHOD 18 3.1 Patient–Image Registration 20 3.2 Hand–Eye Calibration 22 3.3 Accuracy Issue of Two-Step Calibration Method 25 Ⅳ. PROPOSED CALIBRATION METHOD 27 4.1 One-Step Calibration Method 28 4.1.1 Feature Extraction 30 4.1.2 Initial AR-Core Transformations 35 4.1.3 Fast Approach to find Correspondences 38 4.1.4 Refinement 41 4.1.5 Least-Square Method 42 4.2 Summary for Process of One-Step Method 43 4.3 Aspect of Convenience 43 V. EXPERIMENTS 46 5.1 Experiments with Synthetic Data 46 5.1.1 Initialization for Simulation 46 5.1.2 Noise Addition 49 5.2 Experiments with Real Data 50 5.2.1 Phantom Design 50 5.2.2 Experimental Setup 53 5.2.3 Evaluation 57 VI. RESULTS 61 6.1 Results with Synthetic Data 61 6.2 Results with Real Data 79 VII. DISCUSSION 95 VIII. CONCLUTION 99