Improving 3D imaging systems with active pattern illumination

Three-dimensional reconstruction of real-world objects is a crucial task in computer vision and has a wide variety of applications. One example of an industrial application is 3D inspection in production processes, where the 3D geometry of a produced object is compared with its 3D geometry specifications. Here, a particular challenge is to reconstruct 3D models from image acquisitions of objects with homogeneous surfaces without textures. Without available surface textures, feature-based 3D reconstruction approaches fail to recover the geometry of the object. In this thesis, we propose to project a pattern on surfaces to compensate for the absence of visual texture and consequently enable 3D reconstruction of objects without texture. In
particular, two 3D reconstruction approaches covering different domains are studied, a microscopic depth from focus approach and a macroscopic multi-view stereo approach. For both approaches, we analyze the impact of an additional pattern illumination module on the 3D reconstruction quality using synthetic ground truth data as a reference. The qualitative and quantitative results clearly show that the enrichment of textures using a pattern illumination module leads to more accurate
3D reconstructions.