DELAUNAY TRIANGULATION BASED SPARSE 3D RECONSTRUCTION OF INDOOR SCENES USING A MONOCULAR CAMERA by

The role of robots is increasing in our daily life. They can wash dishes, explore unknown terrain, perform security tasks, assist surgeons in patients’ operations, perform surveillance and many more. Human beings gather information about the terrain using eyes and motion of the body. In other words, they see the world in 3D and accumulate information about the structure of the world. In order to approximate some of the capabilities of human beings in robots, we need the ability to create 3D models of the scene, so that robots can navigate smoothly. This technology could be utilized by armed forces, in order to understand the structure of a collapsed building to locate positions of terrorists, or by rescue robots, for example, to understand the structure of a building to locate survivors. In this research study, I make a 3D model of an indoor scene using structure from motion. My method is based on sparse points from a pair of images. My 3D model is based on 2D Delaunay triangulation. I choose this triangulation, because it gives maximum weight to the least angle of all the angles in a triangle. This triangulation usually avoid skinny triangles and is thus most suitable for mapping image texture to a 3D mesh. I render the 3D model using OpenGL.