Affine Camera for 3-D Retinal Surface Reconstruction

Abstract

Abstract. We study 3D retinal surface reconstruction by using an affine camera due to two following reasons: (1) NIH's retinal imaging protocols specify a narrow field of view and (2) each retinal image has small depth variation. Specifically, we incorporate the prior knowledge of human retina geometry in the reconstruction process, and introduce a pointbased approach to estimate the retinal spherical surface. We also show that lens distortion removal and affine bundle adjustment improve the reconstruction error in terms of the deviation from the underling spherical surface. Simulation results on both synthetic data and real images show the effectiveness and robustness of the proposed algorithm.

Automatic Tool Landmark Detection for Stereo Vision in Robot-Assisted Retinal Surgery

“…We are interested in an online stereo system that triangulates and reconstructs surfaces close to real-time from a pair of stereo images. Thus we deviate from the standard calibration methods based on affine reconstruction [4], [11] and triangulate Euclidean shapes directly using calibrated cameras. The affine camera projection is modeled by the projection matrix M ∈ P Affine ⊂ R 2×4 as u = M x 1 .…”

“…In this context we solve two important vision problems online: detecting the tool points accurately in the images and reconstructing the retina and the tool points in scale using the stereo microscope camera. Both are challenging problems on their own [25], [23], [4]. Detecting tool landmark points requires to take into account changes in viewpoint, defocused images, specularities and fast movements of the tool.…”

“…We tackle the problem of stereo microscope calibration using the detected tool landmark points, by assuming a full affine camera model [9] for each microscope camera. Previous methods [4], [11] propose calibration of the affine camera by first reconstructing the object with affine Structure from Motion (SfM) [26] and then computing the suitable upgrade for calibration. This may not be reliable due to the inherent problems of affine factorization-based SfM with respect to noise, the occurrence of missing data, and reconstruction ambiguities.…”

Analysis of shape assumptions in 3D reconstruction of retina from multiple fundus images

“…Although retinal image processing is a well studied field, there are a few studies about 3D reconstruction of retinal fundus [3,4,5,6,7,8]. Liu et al [3] estimate the epipolar geometry and projection matrices after a self calibration.…”

“…The method by Martinez-Perez et al [5] involves an offline calibration stage and provides 3D reconstruction of the retinal vessels. Chanwimaluang et al [6] present a retinal surface reconstruction technique using an affine camera model, since they work on retinal images with small field of view. Lin et al [7] perform a 2D-to-2D registration followed by a multi-stage bundle adjustment.…”