面向大视场高清光场成像的超分辨率三维重建方法研究

For the sake of realizing wide-FOV high-resolution light field imaging, the project designs a novel structure of camera array which capture multiple images simultaneously with lapped and gradually changed fields of view. Consequently, the depth information in 3D space can be obtained together with the requirements of wide-FOV and high-resolution imaging by combining the merits of panoramic mosaicking and super-resolution light field imaging, Specifically, we intend to conduct research on the following three subjects: (1) Through the design and implementation of an experimental prototype of camera array with lapped and gradually changed fields of view, prove the scientificity and feasibility of realizing wide-FOV panoramic imaging and super-resolution 3D reconstruction simultaneously; (2) Due to the existing of small baselines, wide baselines and orthogonal baselines in camera array, propose a multi-view stereo matching algorithm which has high accuracy as well as high precision; (3) Under the consideration of registration uncertainty, propose a unified theoretical framework combining the researches of wide-FOV panoramic mosaicking, super-resolution and 3D reconstruction. Moreover, a solving method of iterative optimization will be presented. In summary, the research not only has an important meaning in basic theories and technologies, but also will greatly motivate the applications of panoramic imaging and light field imaging in the areas of military long-range reconnaissance and high-resolution video surveillance.

本项目通过设计一种新颖的叠进式摄像机阵列结构，将拼接式全景成像技术和超分辨率光场成像技术有机结合，基于多帧视场相互存在重叠又渐进延拓的图像，探索大视场超高清光场成像系统实现的关键技术，在兼顾“看得清”和“看得广”的同时，还要求获取场景的三维深度信息。研究内容：1）通过实验原型系统的设计实现，证明叠进式摄像机阵列结构同时实现大视场全景成像和超分辨率三维重建的科学性和可行性；2）针对摄像机阵列中存在正交基线的特点，结合小基线和大基线的优势，研究实现一种既保证准确度又提高精确度的多视图立体匹配算法；3）在考虑配准不确定性的前提下，结合三维深度信息，提出一种将大视场全景拼接、图像超分辨率和三维重建研究相统一的理论框架，并给出迭代优化的求解方法。研究不仅具有理论创新和技术突破，更重要的是将对全景成像技术和光场成像技术在军事远景侦察、高清视频监控等领域的发展和推广应用带来质的变化和飞跃。

本项目研究提出一种叠进式的摄像机阵列结构设计方法，将拼接式全景成像技术和超分辨率光场成像技术有机结合，基于多帧视场相互存在重叠又渐进延拓的图像，探索大视场超高清光场成像系统实现的关键技术，在兼顾“看得清”和“看得广”的同时，还能获取场景的三维深度信息。研究内容：1）通过实验原型系统的设计实现，证明叠进式摄像机阵列结构同时实现大视场全景成像和超分辨率三维重建的科学性和可行性；2）针对摄像机阵列中存在正交基线的特点，结合小基线和大基线的优势，通过快速特征点提取以及面元的构建、过滤和扩展技术，研究实现一种既保证准确度又提高精确度的多视图三维重建算法；3）在考虑配准不确定性的前提下，结合三维深度信息，提出一种将大视场全景拼接、图像超分辨率和三维重建研究相统一的理论框架，并给出迭代优化的求解方法。研究不仅具有理论创新和技术突破，更重要的是将对全景成像技术和光场成像技术在军事远景侦察、高清视频监控等领域的发展和推广应用带来质的变化和飞跃。