基于虚拟成像原理的组合相机视场拼接与波段配准方法  

A View Stitching and Band Registration Method for Composite Cameras via Virtual Imaging

作　　者：冯笑宇 李雷[1] 钟联宇 陈江苏 顾杭 FENG Xiaoyu;LI Lei;ZHONG Lianyu;CHEN Jiangsu;GU Hang(China Helicopter Research and Development Institute,Tianjin 300480,China)

机构地区：[1]中国直升机设计研究所,天津300480

出　　处：《无线电工程》2024年第6期1576-1583,共8页Radio Engineering

基　　金：国家自然科学基金(6181373)。

摘　　要：针对宽幅组合相机在交会角过小、平差精度不稳定等方面的挑战,以及多光谱组合相机面临的波段配准技术难题,深入探讨了虚拟成像原理在扩大视场角和提升波谱分辨率方面的实现途径。通过分析和优化组合相机各物理相机间固有的几何关系,将多台物理相机的位置和姿态关系视为刚体连接。通过测量不同相机拍摄的影像间的同名像点,解算相机间的几何变换关系并重投影至虚拟成像面,最终实现了各物理相机所获取的独立图像的高效且精确的影像融合,融合的图像符合单中心投影规律。通过优化,扩展了宽幅组合相机的航摄范围,提高了多光谱组合相机的光谱分辨率,尤其在处理无纹理区域和地形起伏场景时展现出卓越的适应性和精确度。实验结果显示,对于宽幅组合相机,该方法的影像拼接精度达到0.31 pixel;对于多光谱组合相机,波段配准精度达到0.25 pixel。证明了虚拟成像技术能够显著提升组合相机数据的处理性能和效率,且与现有摄影测量处理软件系统完全兼容,拓宽了组合相机数据在航空摄影测量领域的应用范围,为地理信息系统的应用提供了新视角。To overcome the challenges posed by the small intersection angle and unstable triangulation precision of large-format composite cameras,as well as the technical difficulties in band registration for multispectral composite cameras,the application of virtual imaging principles is explored in depth to expand the field of view and enhance spectral resolution.The research involves analyzing and optimizing the inherent geometric relationships between the individual physical cameras within a composite camera system,treating the position and attitude relationships of multiple physical cameras as rigid connections.By measuring homologous image points across different camera-captured images,the geometric transformation relationships between cameras are computed,and the images are reprojected onto a virtual imaging plane,ultimately achieving efficient and precise image fusion of independently captured images by each physical camera,in accordance with the principles of central projection.This optimization method not only extends the aerial photography range of large-format composite cameras but also improves the spectral resolution of multispectral composite cameras,particularly demonstrating exceptional adaptability and accuracy in processing textureless regions and terrain undulating scenes.Experimental results show that for large-format composite cameras,the image stitching accuracy of this method reaches 0.31 pixel;for multispectral composite cameras,the band registration accuracy achieves 0.25 pixel.This validates that virtual imaging technology can significantly enhance the processing performance and efficiency of composite camera data,and is fully compatible with existing photogrammetric processing software systems,thus broadening the scope of application of composite camera data in the field of aerial photogrammetry and offering new perspectives for the application of Geographic Information Systems.

关 键 词：虚拟成像 波段配准 宽幅相机 多光谱相机 多拼相机 组合相机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]