Volumetric modeling through fusion of multiple range images with confidence estimate

This thesis describes our investigation of creating 3-D models from multiple range images. There are many aspects to this research. We begin with an evaluation of the occupancy grid approach to fusing range data. After our evaluation, we develop a confidence metric that provides additional information about the workspace. Our research in the occupancy grid method shows that it is not ideal for our application using range images. A review of the volumetric modeling literature leads us to believe that we can develop a new technique that has the advantageous properties of the occupancy grid but is designed to be as efficient with range images as other volumetric approaches.

The second half of this thesis describes the new volumetric technique that we have developed. It presents the theory behind the approach, develops the method for combining information from several views, and tells how we extract a surface to create the final model. Finally we give several examples of our technique on real and synthetic data sets.