基于光场子孔径条纹图像的三维重建关键技术研究

Surface three-dimensional profile reconstruction is the basis of modern industrial manufacture and many commercial applications. With the merits of high accuracy and real-time capability, digital fringe projection profilometry has received extensive concerns from academics and industrials. As an interdisciplinary subject, the techniques and theories in optics, informatics and computer vision have been used in fringe projection profilometry.The quality of the three-dimensional result reconstructed with fringe projection profilometry is always degraded by the high dynamic reflectivity of object surface, since the traditional camera only responds to the intensity of light. The light field camera can record the angular and intensity of light, so that the three-dimensional result is achieved by matching the sub-aperture images captured by light field camera through stereo vision. But the matching of sub-aperture images is always not efficient. This project aims to combine the advantages of fringe projection profilometry and light field camera together to improve the performance of three-dimensional reconstruction. Firstly, the fringe projection scheme suitable for sub-aperture image is proposed. Secondly, the noise model of the wrapped phase of sub-aperture fringe image is established, the strategies to eliminate the noise in wrapped phase are developed correspondingly. Thirdly, the methods to obtain and correct the absolute phase recovered from the wrapped phase of sub-aperture fringe image are developed. Fourthly, the approach to integrate the three-dimensional results reconstructed by multi-sub-aperture fringe images is proposed to avoid the degradation caused by high dynamic reflectivity of object surface. Fifthly, the fringe order and absolute phase are used to guide the matching of sub-aperture fringe images for high-efficiency and high-accuracy stereo vision. The achievements of this project will be implemented to meliorate the performance of three-dimensional reconstruction with sub-aperture fringe images and promot its industrial applications.

表面三维重建是现代工业生产和商业应用的重要需求，数字条纹投影三维重建技术由于结果准确、具有实时性成为国际学术界和产业界关注的焦点，是光学、信息学、计算机视觉领域深度交叉的研究方向。采用只响应光强的传统相机拍摄条纹，条纹投影三维重建结果容易受到物体表面反射特性影响而降低质量。光场相机能记录光线的角度和强度，可利用子孔径图像立体匹配实现三维重建，但效率有待提高。本项目综合两种技术优势，对基于光场子孔径条纹图像的三维重建技术深入研究。首先研究适合光场子孔径图像的条纹投影机制，建立子孔径条纹图像包裹相位的噪声模型及抑制噪声策略，然后提出绝对相位获取和修正方法，再提出子孔径图像三维重建结果融合新方法解决不均匀反射造成的三维重建精度下降问题，以及利用条纹阶数和绝对相位引导子孔径图像实现立体匹配，快速获得重建结果。本项目开展，有利于解决基于光场条纹的三维重建理论问题、促进该技术的产业化应用。

本项目通过探索光场子孔径条纹图像的特点，对基于光场子孔径条纹图像的三维重建关键技术进行了深入探讨。从光场子孔径图像之间的空间几何关系及子孔径条纹信号数学特性出发，得到了光场投影条纹信号的统一表示形式和模型。项目按照原计划进度进行，达到了预定的研究目标。在光场子孔径条纹信号绝对相位获取方面，研究了适用于高动态表面的高可靠度光场子孔径条纹信号绝对相位获取方法；在子孔径三维重建结果融合方面，研究了子孔径图像空间位移估计方法和基于该位移的三维重建结果融合方法。随着研究的深入，根据实际情况和需求，还开展了可靠绝对相位和条纹阶数引导的快速双目立体匹配方法研究。项目具体研究成果有基于时间相位展开的光场子孔径条纹信号绝对相位获取方法；绝对相位和条纹阶数引导的快速双目立体匹配三维重建方法；光场子孔径三维重建结果超分辨率融合方法。与相关文献进行比较，我们提出的方法能够以更高的可靠性、更高的重建精度和较高的效率完成结构照明光场三维重建，为相关技术的实际产业化应用提供了部分技术基础和理论指导。本项目总共培养了10位硕士研究生，拟投稿3篇SCI权威期刊论文和申报国家发明专利3项，超过了立项之初规定的目标。