An approach to generate geometric models from multiple range images

The research described in this dissertation focuses on the development of a new approach for the generation of geometric models from multiple-view range image data.;Through intensive comparison and evaluation of different representations, the cross-section contour based representation is concluded to be ideal for modeling with range image data. The representation is shown to be at an intermediate level--compatible with both the low-level of range image data and with the need to provide relatively high-level geometric and topological information in models.;A new concept of generating partial models within device frames, frames associated with the working principle and geometry of a range sensor, is introduced. The range data are well distributed in the device frame. This good data distribution facilitates computations relevant to rendering the cross-sections required by the representation and relevant to identifying occlusions present in the image. Methodology for merging the partial models with a current global model is developed to allow the incorporation of redundancy between the partial model and the current global model and to allow growth of the global model. A simulation of the ERIM imaging-radar based range sensor, a prototype triangulation-based range sensor developed for this research and a commercial HYMARC range sensing system are used for approach verification. The device frames associated with the sensors are derived, and used to test the modeling approaches and the developed system.;The presented research: demonstrates the suitability of the cross-section based representation for range-image based modeling systems; introduces a new concept and associated methods for generating cross-section contour models in range sensor device frames to take advantage of well distributed data; develops a series of algorithms for partial modeling in the device frame and for global model integration; and demonstrates the feasibility of the developed new approaches for applications by testing the system for multiple sensor types.