Research On Key Issues In Structured Light Based 3D Sensing System

3D sensing technology, as a significant extension of human sensing capability for external environments, plays an important role in a variety of disciplines, such as humancomputer interaction, medical imaging, manufacturing, 3D printing, remote sensing mapping, intelligent monitoring, cultural relics preservation and so on. In some applications, the existing 3D sensing technologies have achieved good results, however, because of some restricts in these technologies, better performance applications are blocked. In this thesis, we research on some key issues in the structured lighted based 3D sensing system, such as the density of point cloud in one-shot 3D sensing, high accuracy 3D sensing for dynamic scenes and real time accurate 3D sensing, and propose corresponding resolutions.Firstly, for dense one-shot 3D sensing, a stereoscopic fringe analysis method is proposed. This method takes advantage of two conventional techniques, stereo and Fourier fringe analysis(FFA). While FFA is competent for high-density and high-precision phase measurement, stereo solves the phase ambiguity caused by the periodicity of the fringe. By jointly using the intensity images and unwrapped phase maps from stereo, the pixel-wise absolute depth can be obtained through a sparse matching process efficiently and reliably. A dense one-shot 3D sensing system is designed based on above methods, and the experiments for real scenes verified the availability of proposed methods. Due to its singleshot property and low complexity, the proposed method facilitates dense and accurate 3D sensing in time critical applications.Secondly, for accurate 3D sensing of dynamic scenes, we propose a Fourier assisted phase shifting method, basing on 3-frame phase shifting fringe analysis. By adding motion estimation, we reconstruct the model of 3-frame phase shifting method for dynamic scenes. Through a series of accuracy analysis, the availability of the method is proved, and a scheme for better motion estimation is proposed. Finally, we build a dynamic 3D sensing system basing on the proposed method, and conduct 3D reconstruction for objects in different motion styles. From the quantitative and qualitative evaluation, we can confirm that the proposed method can achieve as good performance as previous best method for uniform motion, and better performance than the previous best method for non-uniform motion.Thirdly, for real time accurate 3D sensing system, an efficient parallel spatial phase unwrapping method is proposed, following the above work for accurate 3D sensing of dynamic scenes. By embedding a sparse set of markers in the fringe patterns, we realize fast parallel absolute phase unwrapping for multi objects and complex surfaces without increasing more frames, consequently avoid affecting the motion applicability of FAPS. Combining FAPS and parallel spatial phase unwrapping, a real time accurate 3D sensing system is built, and the depth map generation speed is up to 30 fps after GPU accelerating. Finally, the high accuracy and high reliability of the real time system is verified by a real time 3D reconstruction experiment for a complex scene consisted of both dynamic and still objects.