A complete list of publications associated with this project can be found on the project's RI web page.

Dissertation

Automatic Three-dimensional Modeling from Reality - The most detailed version of the work. Includes many experiments that are unpublished elsewhere (slides from the defense).

Thesis proposal

Automatic 3D modeling from range images (unpublished) - An early version of the work, this document lays out the main ideas behind automatic modeling and and shows that the idea is feasible (slides from the presentation).

Other automatic modeling publications

Fully Automatic Registration of Multiple 3D Data Sets (IVC 2003) - Journal version of the CVPR 2001 workshop paper combined with the 3DIM 2001 paper. It includes additional experiments and results.

Automatic 3D underground mine mapping (FSR 2003) - An example application of automatic modeling.


3D Modeling Using a Statistical Sensor Model and Stochastic Search (CVPR 2003) - Describes an improved version of the sensor modeling technique for evaluating model hypotheses and a new stochastic search method for finding the best hypothesis (poster part a, part b).

Fully Automatic Registration of Multiple 3D Data Sets (CVPR 2001 workshop) - Extends the initial modeling algorithm by integrating n-view registration refinement into the model construction process and compares the performance of three sequential model construction algorithms (slides from the presentation).


Automatic 3D Modeling Using Range Images Obtained from Unknown Viewpoints (3DIM 2001) - First published paper on automatic modeling. Introduces the model graph, compares several surface consistency measures, and demonstrates a sequential model construction algorithm based on minimum spanning trees (slides from the presentation).

My research on automatic modeling arose from earlier related work on terrain modeling (the most relevant paper is A New Approach to 3-D Terrain Mapping). The automatic modeling and terrain mapping projects are built upon the surface-matching and object recognition research of Andrew Johnson as part of the ARTISAN project.

Next: Video