A new technique has been proposed to generate adaptive triangular meshes from large dense range images. It consists of two stages. The first stage (adaptive sub-sampling) computes a compact triangular mesh by means of both an adaptive sampling of a user defined number of points and a data-dependent triangulation technique. Thus, the shapes present in the original range image are approximated by reducing redundant information. The second stage (error-based decimation) removes nodes of the generated triangular mesh according to an approximation error criterion. In order to speed up this second stage, an estimation of the approximation error is used. Below, two images illustrate each stage.
A. D. Sappa, M. Devy and M. A. Garcia, ``Modeling Built Environments from Large Range Images", 8th Int. Symp. on Intelligent Robotic Systems, pp. 23-29, Reading, UK, July 2000.
J. Bozier, M. Devy and A. D. Sappa, ``A Geometrical Approach for the Incremental Modelling of Free Form Surface by Triangular Meshes", 8th Int. Symp. on Intelligent Robotic Systems, pp. 13-21, Reading, UK, July 2000.
The first image shows the adaptive subsampling and the second image the error-based decimation of a triangular mesh approximating a panoramic range image defined by 1400x8000 points.
