From stereoscopic images to semi-regular meshes

• Direct semi-regular reconstruction from stereo to simplify the classical pipeline.
• Based on a hybrid Poisson-disk sampling, and a coarse-to-fine refinement process.
• Restrict the use of 3D information, and work on the stereo images.
• Efforts are made to reduce as much as possible the runtime using GPU implementations.
• Runtimes are faster than prior semi-regular reconstruction method.

The pipeline to get the semi-regular mesh of a specific physical object is long and fastidious: physical acquisition (creating a dense point cloud), cleaning/meshing (creating an irregular triangle mesh), and semi-regular remeshing. Moreover, these three stages are generally independent, and processed successively by different tools. To overcome this issue, we propose in this paper a new framework to design semi-regular meshes directly from stereoscopic images. Our semi-regular reconstruction technique first creates a base mesh by using a feature-preserving sampling on the stereoscopic images. Afterwards, this base mesh is passed to a coarse-to-fine meshing process to get the semi-regular mesh of the original surface. Experimental results prove the reliability and the accuracy of our approach in terms of shape fidelity, compactness, but also runtime, since many steps have been parallelized on the GPU.

Figure optionsDownload as PowerPoint slide