Detection of Point Correspondences for 3D Reconstruction in Panorama Endoscopy

Abstract: A geometric filter method is developed for the robust detection of point correspondences in endoscope images acquired with the axial motion stereo geometry. Displacements of the detected point correspondences due to system deviations from the axial stereo geometry are analysed and corrected. A comprehensive evaluation including a 3D reconstruction of simulated image sequences, a 3D reconstruction of bore holes from metal parts including camera calibration, and a developed method for estimating parameters for image dewarping prove the robustness and precision of the point correspondences even for weakly structured surfaces. The reconstruction of the bore holes shows very promising results with a at least ten times higher precision compared to preceding approaches.