基于复眼的多孔径合成大视场立体成像技术研究

Wide field of view(FOV) stereo imaging has been widely used in our daily life, scientific research, military and other field, and it is in great demand. Traditional catadioptric omnidirectional stereo imaging system exists field blind in the direction of optical axis, and the formed image has large distortion and low resolution. Fisheye lens stereo imaging system has serious barrel distortion which is difficult to correct, and the resolution of the system is also poor. Compared with the imaging methods mentioned above, compound eye is a more effective way to achieve wide FOV imaging because of wider FOV, smaller distortion, and higher resolution, and therefore this project focusing on the research of the wide FOV stereo imaging techniques by multi-aperture synthesis based on compound eyes, aims at establishing new wide FOV stereo imaging technology based on multi-aperture stereo-field segmentation and synthesis. The technique has the advantage not only in wide FOV imaging but also giving a scene depth information. The project uses the method of multi-aperture stereo-field segmentation to solve the problem of stereo maps caused by non-single viewpoint in the proceeding of compound eyes imaging. Step calibration method is used to calibrate multi-aperture wide FOV system. Multi-aperture depth of integration method is applied to obtain wide FOV depth information of the scene based on boundary segmentation. Implementation of this project could provide new thoughts and solutions for the design of wide FOV stereo imaging system, and lay theoretical and technical basis for the subsequent application of wide FOV stereo imaging.

大视场立体成像技术在日常生活、科研及军事等领域的应用和需求越来越广泛。传统的折反射全向立体成像系统在光轴方向存在视场盲区，并且图像畸变大、分辨率较低；鱼眼镜头立体成像系统会产生严重的桶形畸变且很难校正，成像分辨率较低。与上述成像方式相比，仿生复眼具有视场更大、畸变更小和分辨率更高等优点，因而本项目致力于“基于复眼的合成孔径大视场立体成像技术”的研究，目的是建立一种基于复眼的多孔径立体视场分割与合成的大视场立体成像技术，它既保留复眼大视场成像的优点，又能给出大视场内场景的深度信息。本项目使用多孔径立体视场分割的方法，解决复眼成像非单一视点所造成的立体映射问题；通过分步标定方法，实现多孔径大视场系统的标定；采用基于边界分割的多孔径深度融合，获得大视场场景的深度信息。本项目的实施可为大视场立体成像系统的设计提供新的思路和解决方法，为后续的应用奠定理论和技术基础。

大视场立体成像技术在日常生活、科研及军事等领域的应用和需求越来越广泛。与传统的成像方式相比，仿生复眼具有视场更大、畸变更小和分辨率更高等优点，可满足人们对广域高敏感检测的需求。目前仿生复眼研究成果已成功用于机器人环境感知、制导武器和视频监控等领域。.本项目为探索复眼在合成孔径大视场立体成像中的应用，在研究过程中，实现了基于复眼的合成孔径大视场立体成像技术，并完成与其相关的光学设计、系统标定和合成孔径算法等方面的工作，为大视场复眼立体成像技术的进一步发展提供理论支持。本项目所做的工作如下：.1）模仿昆虫复眼的成像机理，提出了一种基于摄像机阵列的仿生复眼结构设计方法，采用摄像机的矩形视场对预定视场进行无缝划分，根据视场与摄像机的映射关系确定摄像机的排列方式。.2）复眼大视场标定时，利用传统方法很难获取清晰的图像。项目组提出了基于相位标靶的离焦摄像机标定方法，所设计的相位标靶包括圆光栅阵列、契形光栅阵列和彩色圆光栅阵列，均表现出较强的抗模糊性，适用于离焦摄像机的标定。.3）利用相位标靶离焦性能好的优点，将其应用于多摄像机的全局标定，精确确定了外层摄像机相对于中心摄像机的旋转平移矩阵。.4）采用半全局匹配SGM算法，计算双目摄像机的立体视差图；根据原始图配准求得透视矩阵，将不同视场的视差图映射拼接，获得全景立体视差图，实现全景立体成像。.5）参考传统双平面标定法，使用一维标靶和高精度转台来实现虚拟的圆柱面标靶，实现了复眼全视场内的自动化标定。建立了一套完备的圆柱面搭建实验对准流程，使标定具有较高的精度和较低的系统误差。.6）提出带权重的空间多光线交汇算法，并对整个复眼视场内的探测精度进行了评估。推导了基于最小二乘法的空间多光线相交算法，阐述了子眼通道的识别原理，通过探测多个圆柱面点集和一个平面点集对复眼的三维探测精度进行了全方位的评估。