| Image registration is a core task for various applications, such as digital photogrammetry, computer vision, remote sensing, vision-aided navigation and medical imaging. It is a method for estimating geometric transformations of images from an assortment of correspondences between images acquired at different times, perspectives, or even by different sensors. As an increasing number of Earth observation image sensors provide multiple image coverage worldwide, the need for registering imagery acquired from different airborne and space borne platforms is growing rapidly.;This dissertation is focused on developing a robust registration technique between LiDAR (Light Detection And Ranging) and optical imagery (aerial and satellite images). Registering LiDAR intensity and optical images is an especially difficult task due to their substantially different characteristics, such as different sensing methodologies (e.g. wavelength, passive/active image acquisition) and geometric and radiometric differences. Reviewing and testing popular multiple domain image registration techniques, such as feature-based SIFT (Scale Invariant Feature Transform), intensity-based MI (Mutual Information) method, and frequency-based LPFFT (Log-Polar Fast Fourier Transform), it is realized that no single technique could solve the LiDAR/optical image registration completely. Alternatively, a new approach to robust LiDAR/optical image registration, taking advantages of feature-, intensity- and frequency-based methods, is proposed. Using our somewhat limited datasets, the proposed method showed good performance, achieving pixel-level registration precision.;The proposed method is also applied to a specific application -- to generate digital bridge model by fusing LiDAR and aerial imagery. Although the concept of fusing LiDAR and aerial image data to create digital models of man-made objects is not new, our approach is quite different from others due to the novelty of our registration method.;Reviewing the course of this research in its entirety and analyzing results by applying the new LiDAR/optical image registration approach on our datasets, we can state that it is a novel contribution to the growing multiple domain image registration field. |
