融合机载多视倾斜影像和LIDAR数据的城市"真"正射影像自动纠正方法研究

Digital orthophotos are a critical component of the National Spatial Data Infrastructure. Because only the geometric distortion caused by imaging process and terrains are considered, the conventional orthorectification of large-scale aerial images has encountered many difficulties, e.g., occlusions problem as a result of man-made buildings, shadow problem due to different solar angles, etc., in many cases, especially in urban area. Hence, there are increasing efforts toward the generation of "True" orthophotos for not only representing the actual coordinate of the region but also providing useful information of the region, although it was subjected to several challenges in practice: a)The generation of high-quality DSM for image geometry ortho-rectification is ultra time-consuming and costly,especially when all kind of rooftops are modeled in the correct way; b)Occlusions occurred in the single images have to be filled up by combining several overlapping orthophotos. But often not all occluded areas can be filled up as not enough overlapping orthophotos are available; c)Shadows from high buildings are popular in urbar area. But often the removal of various shadows accurately and in detail can be complicated if computed automatically or time consuming if identified interactively; d)Other issues,like seamless mosaic, color balancing, etc..This project intends to provide one economically viable solution to generate urban "true" orthophotos by fusing airbore oblique multi-view images and LIDAR data. Several key techniques are to be developed: ①high-precision registration of oblique multi-view aerial images with LIDAR data by integrating the Laplace spectrum matching and photogrammetry AAT techniques; ②generation of high-quality urban DSM through Morphological grayscale geodetic reconstruction from LIDAR data and point cloud "boundary" reconstruction based on multi-vision image constrains; ③a series of large-scale aerial images "true" orthorectification techniques like automatic building occlusion detection and re-filling based on the difference of orthorectified multi-vision aerial images, automatic building shadow detection and enhancement based on the conformity of orthorectified multi-vision aerial images and seamless mosaic of decorated building region based on smoothed "step curve" are presented. This project will promote to generate "true" orthophotos for large-range urban region effectively and efficiently and provide accurate geographical information for urban GIS modeling or other remote applications.

成本高昂、周期长是目前城区大比例尺"真"正射影像制作面临的"瓶颈"问题。本研究以机载LIDAR与倾斜摄影系统获取的"点云"、多视觉倾斜影像为数据基础，就城区"真"正射影像自动纠正方法展开研究：第一，提出和实施基于谱空间匹配与自动空三联合处理的三维点集整体匹配策略与方法，实现LIDAR"点云"与多视倾斜影像的精确地理配准；第二，发展顾及对偶特性的机载Lidar数据形态学滤波方法，并通过多视觉约束下的"点云"边界重构过程以及3D/2D混合三角构网技术实现城市高质量数字地面模型的高效生成；第三，提出和实现包括多视觉相异性下的影像建筑遮挡分析与填补、多视觉相同性下的影像建筑阴影检测与增强、多视影像投影差全面改正以及基于"刀刃"曲线的"伪"区域边界无缝镶嵌等一系列的"真"正射影像自动纠正方法与技术，从而实现正射影像精确、真实地理信息的高效获取，为城市基础地理空间信息建设与广泛应用奠定理论与应用基础.

融合多视觉倾斜影像和Lidar数据对城区大比例尺”真”正射影像智能处理理论与技术进行发展、创新研究，着力解决其制作成本高昂、周期长等“瓶颈”问题，完成研究工作如下：①Lidar点云滤波与影像空间配准。融合形态学灰度重建与不规则三角网模型建立滤波框架，有效提高Lidar点云滤波质量与效率；基于点云物方信息密集及非线性光谱差异下的影像相似互信息度量优势，提出和实施“针孔”模拟成像与空间后方交会联合处理下的点集整体匹配策略与方法，成功实现LIDAR“点云”与多视倾斜影像的精确地理配准。②倾斜影像空三处理与DSM高质量匹配生成。改进SIFT仿射不变性能以解决制约倾斜航空影像空三处理效率的影像连接点自动生成与转点问题，融合改进的SIFT和SGM算子以解决制约DSM倾斜航空影像自动生成的密集匹配质量问题，并基于投影分析对多视DSM进行融合，实现城区高质量DSM多视倾斜航空影像自动生成。③城区“真“正射影像制作。以阴影同质区正确选取为突破口提出一种结合全局光照补偿与局部线性变换的光学遥感影像阴影自然重建新方法，基于遮挡导致立体影像匹配错误这一事实引入遮挡约束从错误匹配像素中探测影像遮挡区域，从而有效解决“真“正射影像制作中的阴影信息合理补偿与遮挡区域自动检测两大关键技术问题。④倾斜成像系统研制与相机标定。基于多台普通数码相机研制经济型机载倾斜摄影成像原型系统,并构建全自动数控相机检校平台实现非量测相机的高精度、高效率标定，为我国倾斜航空摄影技术的广泛应用以及产业化推动奠定坚实的理论与技术基础。