HOMEABOUTPEOPLEPROJECTSPUBLICATIONSHISTORYIMAGES gdc

Augmented Reality - ARToolkit Patternmaker


Goal

To automatically create large numbers of target patterns for the ARToolkit.

Approach

The ARToolkit normally requires custom drawn target patterns that are hand trained using the toolkit and a camera. We want the capability of generating large numbers of target patterns and training them automatically. Furthermore, we want the patterns to be as clear as possible to provide the highest performance recognition in the toolkit.

Participants

Results

We have created the patternMaker program, which generates grids of black and white squares. The restriction is that the grids cannot be rotationally symmetric with itself or other grids in the group, else the ARToolkit won't be able to extract orientation from the target pattern. The training files are automatically generated from the patterns.

There are two ways to use this release:

Compared to the sample patterns provided with the ARToolkit, we have found that our patterns can be recognized at much greater distances or are much more stable when close. In some sense, the large black and white squares are optimal targets. We have used a 4x4 target because the ARToolkit uses a 16x16 grid internally, and the 4x4 target maps more cleanly than a 3x3 pattern does even though the 3x3 should be visible at larger distances. Modifying the ARToolkit should make other pattern sizes more effective in the future.

Download

The example patterns and support patternMaker software can be downloaded at our download page.

Support

This project is supported by the NSF STC for Computer Graphics and Visualization and collaborative NSF/NIH projects with the Scripps Research Institute and the University of Washington. The ARToolkit is publicly available from the University of Washington.



Last update: July 5, 2002