基于重影成像模型的单帧图像三维重建技术及其应用研究

Camera imaging is a process of projecting a three-dimensional scene onto a two-dimensional image plane; this process would lose a lot of information. Since perspective projection is irreversible, a single image three-dimensional reconstruction without constraints is an ill-posed problem, there is no general solution. This study presents a novel solution to the problem of three-dimensional reconstruction from a single image. We address the problem by proposing a novel depth-dependent imaging model called echo imaging model. In our method, by coding the projection light rays, the depth information of scene is embedded in image. For solve the hard problem of three-dimensional reconstruction from single image, we research single view 3D reconstruction techniques according to the proposed echo imaging model, full-focus image generation techniques, and refocus technology of echo image. Based on a theoretical analysis, the echo projection model is equivalent to a multi-view stereo vision model, in which a multi-baseline 3D reconstruction algorithm can be achieved for improving the robustness of the three-dimensional reconstruction results. At the same time, in the study we also propose a design guideline and an optimization method of light encoder for imaging echo image, and accordingly design an ultra-small device with large FOV for three-dimensional reconstruction, that will be applied to three-dimensional capsule endoscope system to estimate the internal shape of the human organs. The research is a further development of technologies for computational imaging, and can improve the quality of the single image 3D reconstruction.

图像是三维空间在二维空间的投影。投影过程的不可逆性决定了无约束条件下单帧图像三维重建过程是一个病态问题，没有一般解。本项目着眼于单帧图像三维重建难题的成因- - -单帧透视投影图像深度信息不完备问题，探索新的深度依赖成像模型- - 重影成像模型，通过编码投影光线在图像中嵌入场景深度信息，提升静态图像中深度信息的完备性；在此基础上研究基于重影成像模型的单帧图像三维重建技术、全聚焦图像生成技术和图像焦点重置技术，从根本上解决单帧图像三维重建难题。理论分析表明重影成像模型与多视立体视觉模型等价，可以在单帧图像上实现多基线三维重建算法，能够提高三维重建结果的鲁棒性。同时，本项目给出针对三维重建目标的光线编码器设计指针与优化方法，并据此设计一个超小型、大视场三维重建系统方案，应用于三维胶囊型内窥系统，实现人体器官腔体形状的三维重建。通过本项目研究，将进一步完善计算成像理论，大幅度提升单帧图像三维重建效果。

图像是三维空间在二维空间的投影。投影过程的不可逆性决定了无约束条件下单帧图像三维重建过程是一个病态问题，没有一般解。本项目着眼于单帧图像三维重建难题的成因---单帧透视投影图像深度信息不完备问题，探索了新的深度依赖成像模型--重影成像模型，通过编码投影光线在图像中嵌入场景深度信息，提升静态图像中深度信息的完备性；在此基础上研究了基于多透视投影图像的单帧图像三维重建技术、全聚焦图像生成技术，力求解决单帧图像三维重建难题。理论分析表明多透视成像模型与多视立体视觉模型等价，可以在单帧图像上实现多基线三维重建算法，能够提高三维重建结果的鲁棒性。同时，本项目给出了针对三维重建目标的光线编码器设计指针与优化方法，并据此设计了一个超小型、大视场三维重建系统方案，拟应用于三维胶囊型内窥系统，实现人体器官腔体形状的三维重建。通过本项目研究，进一步完善计算成像理论，解决了单帧图像三维重建问题，提升了单帧图像三维重建效果。