Three-dimensional surface reconstruction using Delaunay triangulation in the image plane

In this paper a technique for 3D surface reconstruction from stereo segments by using a constrained 2D Delaunay triangulation in the image plane is described. The proposed approach provides an appropriate tessellation of stereo data, being consistent with the visibility constraints of the imaging system. It results in a scene surface representation suitable for various applications of computer vision, as 3D object recognition and obstacle detection. Results and comparisons between our approach and Delaunay tetrahedralization are illustrated on real data together with some considerations about computational complexity and running time efficiency.