Structured lighting stereoscopy with marching pseudo-random patterns

Abstract

This thesis discusses the development of a 3D reconstruction system based on active stereoscopy with structured lighting. The proposed system accomplishes 3D reconstruction of objects using a combination of spatial-neighboring and time-multiplexing structured light patterns encoded with pseudo-random codes. This combination builds upon the concept of marching patterns that can adaptively increase the reconstruction resolution on demand. Original techniques are introduced to recover and validate pseudo-random codes from images using a combination of image processing algorithms. The validation of codes is performed such that potential outliers are minimized and the density of 3D point clouds remains fairly high in the models. Finally, the system simultaneously provides models with color and texture mapping using linear interpolation between vertices which enables a more accurate and intuitive representation of 3D objects. Experimental results presented demonstrate the various reconstruction densities possible with the proposed approach, its performance on objects with different surface properties, and the quality of the colored 3D models achieved.