3D-Modelling of seafloor structures from ROV-based video sequences

在水下成像和水下3D重建中，大多数研究人员仍然使用透视相机模型，即使在水下外壳的玻璃界面处发生折射[SK11]。可以很容易地表明，由于折射效应，单视点相机模型（SVP模型）是无效的。尽管这已经知道了一段时间，但没有运动恢复结构系统（SfM系统）明确地对折射效应进行建模。因此，本项目的总体目标是评估折射成像的使用，并将其与经典的透视SfM方法进行比较。初步研究和我们第一个项目阶段的结果已经表明，当重建水下图像序列的相机路径时，使用透视相机模型引起的系统模型误差会导致姿态估计误差[SK 11]。我们期望折射成像的正确物理建模将导致改进的相机模型，从而消除透视3D建模中的系统误差。由此产生的三维模型可用于改进海底结构的体积测量。

In underwater imaging and underwater 3D reconstruction, most researchers still use the perspective camera model, even though refraction at the glass interface of the underwater housing occurs [SK11]. It can easily be shown, that the single-view-point camera model (SVP model) is invalid due to refractive effects. Even though this has been known for a while, no Structure-from-Motion system (SfM system) explicitly models refraction effects. Therefore, the overall goal of this project is to evaluate the use of refractive image formation and to compare it to the classical perspective SfM approaches. Preliminary studies and our results of the first project phase already indicated that the systematic model error caused by using the perspective camera model leads to a pose estimation error, when reconstructing the camera path of an underwater image sequence [SK11]. We expect that the correct physical modeling of refractive image formation will lead to improved camera models that will eliminate the systematic errors in perspective 3D modeling. The resulting 3D models can then be used for improved volumetric measurements of seafloor structures.