Abstract
This thesis addresses the problem of recovering the 3-D structure of a rigid terrain and the trajectory of a panoramic vision system from flyover imagery. It is primarily targeted as an efficient and accurate technology for the mapping of benthic habitats by the deployment of unmanned underwater remotely operated vehicles that can navigate automatically (autonomously) based on visual cues. For this or similar applications, e.g., deployment of unmanned airborne vehicles, an adjustable slanted down-look configuration, producing the so-called conical panorama, is a flexible design to control the scene coverage. A multi-camera realization with overlapping field of views, so-called stereo conical imaging, is analyzed in details. The system model is given by generalizing the pin-hole camera model. Two major investigations are carried out to analyze the conical imaging: (1) performance assessment within a general framework applicable in N-Ocular vision; (2) 3-D super-resolution mosaicking technique to enable the generation of composite 3-D reaps. The qualitative results supported by theoretical findings have demonstrated that panoramic conical imaging is very promising in terms of achievable accuracy in vision-based positioning and 3-D mapping. In addition to numerous experiments with synthesized data and images recoded in air, the system performance has also been tested with underwater images acquired under different turbidity conditions.The investigation has primarily emphasized the geometric issues in the 3-D reconstruction of shape and motion from stereo conical views. For underwater applications, various factors have a significant impact on the radiometric aspects of the recorded imagery. Therefore, incorporating underwater image formation models has been identified as one of key future research directions to enhance the system performance.