This research develops and improves the fundamental mathematical approaches and techniques required to relate imagery and imagery derived multimodal products in 3D. Image registration, in a 2D sense, will always be limited by the 3D effects of viewing geometry on the target. Therefore, effects such as occlusion, parallax, shadowing, and terrain/building elevation can often be mitigated with even a modest amounts of 3D target modeling. Additionally, the imaged scene may appear radically different based on the sensed modality of interest; this is evident from the differences in visible, infrared, polarimetric, and radar imagery of the same site. This thesis develops a `model-centric' approach to relating multimodal imagery in a 3D environment. By correctly modeling a site of interest, both geometrically and physically, it is possible to remove/mitigate some of the most difficult challenges associated with multimodal image registration. In order to accomplish this feat, the mathematical framework necessary to relate imagery to geometric models is thoroughly examined. Since geometric models may need to be generated to apply this `model-centric' approach, this research develops methods to derive 3D models from imagery and LIDAR data. Of critical note, is the implementation of complimentary techniques for relating multimodal imagery that utilize the geometric model in concert with physics based modeling to simulate scene appearance under diverse imaging scenarios. Finally, the often neglected final phase of mapping localized image registration results back to the world coordinate system model for final data archival are addressed. In short, once a target site is properly modeled, both geometrically and physically, it is possible to orient the 3D model to the same viewing perspective as a captured image to enable proper registration. If done accurately, the synthetic model's physical appearance can simulate the imaged modality of interest while simultaneously removing the 3-D ambiguity between the model and the captured image. Once registered, the captured image can then be archived as a texture map on the geometric site model. In this way, the 3D information that was lost when the image was acquired can be regained and properly related with other datasets for data fusion and analysis.
Library of Congress Subject Headings
Image processing--Digital techniques; Multisensor data fusion; Signal processing; Three-dimensional display systems
Department, Program, or Center
Chester F. Carlson Center for Imaging Science (COS)
Walli, Karl C., "Relating Multimodal Imagery Data in 3D" (2010). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus