This dissertation addresses the problem of scanning the geometry of real objects and building accurate computer models of those objects. We present a complete system which employs a structured light scanner to acquire 3D views of objects from multiple viewpoints. These multiple views, expressed in a sensor-oriented coordinate system, are then registered into a model-centred coordinate system, before being integrated into a single mesh describing the object's geometry. Line of sight constraints forbid any single view from capturing the entire surface of an object, so multiple scans must be performed. We have developed registration techniques which may register all of the views simultaneously, resulting in a globally optimal solution. Statistical error modeling of the sensor, and the use of these models in the registration process, forms a key part of the research. It is motiviated by the observation that all measurements are subject to some degree of random measurement error. The true values of these errors cannot be determined, however their statistical properties may be modeled. Our registration system utilises these error models to improve registration accuracy, and to allow the accuracy of the registration to be estimated. The resulting system is a flexible platform for 3D data capture and modeling. It may be used in conjunction with the structured light scanner, or 3D data acquired from any other source. We demonstrate this capability with models constructed from sources such as laser range finders and scanning touch probe systems. The contributions of this thesis are as follows: a novel stereo matching algorithm which permits the estimation of stereo disparity as well as the uncertainty in the disparity, development of a practical 3D vision sensor based on structured light techniques, two novel algorithms for performing simultaneous multiple view point set registration, while supporting individual point error models and estimating the uncertainty in the registration solution, a novel algorithm for efficiently solving the multiple view registration problem, and the implementation of a number of existing surface correspondence and reconstruction techniques, permitting the development of an integrated 3D vision system for capturing and modeling 3D objects.
| Identifer | oai:union.ndltd.org:ADTP/264771 |
| Date | January 2001 |
| Creators | Williams, John Alan |
| Publisher | Queensland University of Technology |
| Source Sets | Australiasian Digital Theses Program |
| Detected Language | English |
| Rights | Copyright John Alan Williams |
Page generated in 0.0021 seconds