A System for 3D Shape Estimation and Texture Extraction via Structured Light

Abstract

Shape estimation is a crucial problem in the fields of computer vision, robotics and engineering. This thesis explores a shape from structured light (SFSL) approach using a pyramidal laser projector, and the application of texture extraction. The specific SFSL system is chosen for its hardware simplicity, and efficient software. The shape estimation system is capable of estimating the 3D shape of both static and dynamic objects by relying on a fixed pattern. In order to eliminate the need for precision hardware alignment and to remove human error, novel calibration schemes were developed. In addition, selecting appropriate system geometry reduces the typical correspondence problem to that of a labeling problem. Simulations and experiments verify the effectiveness of the built system. Finally, we perform texture extraction by interpolating and resampling sparse range estimates, and subsequently flattening the 3D triangulated graph into a 2D triangulated graph via graph and manifold methods.