Conceptual modeling and rendering techniques for architectural design.

Master's thesis, Cornell University, 1998.

Today's design professions are not integrating the computer into their workplace as successfully as more technical professionals. It is tempting to attribute this to the fact that artists and architects are not as technically savvy as engineers and scientists. Closer inspection reveals, however, that current software applications simply do not fit their needs.

Three problematic areas are identified in this thesis in current computer design software. (1) Human-computer interfaces are too cumbersome for design. (2) The current human-computer interfaces do not facilitate transitioning between conceptual drawings and computer models. (3) Current computer rendering styles are not abstract enough for visualizing incomplete designs. These shortcomings are alleviated here by introducing new human-computer interface and sketch rendering The computer graphics community refers to computer images that approximate hand drawings as non-photorealistic images. Because this term describes what the images are not, rather than what they are, the author considers this inappropriate terminology. Consequently, the term sketch rendering is introduced. techniques. A solid modeler was written for this thesis to address the human-computer interface issues. Currently, architects and designers use traditional media - e.g., clay, chip board, and sketches - in preliminary design. For a solid modeler to compete with these materials, it must be as simple to use. Our modeler's interface approaches this goal by incorporating a surface cursor that feels its way along the model's surface, giving the designer much needed visual cues as to the cursor's location in three dimensions.

The need for more abstract rendering styles is addressed by presenting a new technique for generating sketch renderings. These renderings approximate the look and feel of conventional hand drawings. The algorithms presented here simulate an artist's hatch markings with vectored streamlines. The streamlines are calculated in a two-dimensional vector field generated by projecting three-dimensional surface flow onto the image plane. A global illumination image is used as a tone map to control the hatching density. Because the algorithms operate on strictly two-dimensional data extracted from the three-dimensional model, they are robust and simple to implement. Several new support algorithms, including a new one-pass streamline algorithm that approximates both density and direction, are described.