|M.Sc Student||Bashar Elnashef|
|Subject||Linear and Refraction-Invariant Models for Image-Based|
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Filin Sagi|
Underwater photogrammetry benefits a wide variety of applications including seabed mapping, archeological surveys, marine biology, and fluid flow measurement, to name only a few. Nonetheless, development of underwater photogrammetry poses modeling problems. Refraction effect, the extension of the imaging system into a unit composed of both camera and a protecting housing device, lack of fixed reference frames, and imaging from close ranges and in various scales impose significant constraints to the traditional photogrammetric mapping .
To handle these effects, we derive in this thesis a refraction-invariant representation and show that despite the pronounced non-linear and depth-dependent distortions, such a model is attainable. We also show that its contribution is not only theoretical, as it also allows to estimate the pose parameters linearly and at a significantly improved level of accuracy. The thesis then extends the model to calibrate the underwater-related system parameters and, again, shows here the ability to yield a linear model, to simplify the settings and requirements for calibration procedures, and most importantly to improve the estimates of the system parameters by order of magnitude or more. Experiments show not only enhanced parameters accuracy but also stability of the model in the presence of high level of noise. Finally, we also propose a multi-layer model, which can be generalized to a number of layers with different indices of refraction. Differing than existing models, ours exhibits high stability and improved estimation accuracies. Thus, the thesis provides an in-depth look into the geometrical modeling of underwater images and at the same time shows practical enhancement in the accuracies and requirements.