地表三维模型和航拍全景图像在交通建设、水利工程、生态研究、城市规划以及安全防卫等诸多领域能够发挥重要作用，长期以来，研究人员对地表三维模型的生成和航拍全景图像的制备进行了深入的研究。传统上，人们使用固定翼飞机或卫星携带激光扫描设备LiDAR、合成孔径雷达SAR 等装置获取地表的数字高程模型（Digital Elevation Model, DEM），以之作为地表三维模型使用，航拍全景图像则由天基平台采集得到的航空影像拼接所得。 近年来，无人机技术由军用向民用扩散，小型无人机逐渐成为航空影像的一种新型采集工具。相较于固定翼飞机或卫星，无人机具有明显的优势，其体积小、重量轻、携带方便。旋翼式垂直起降无人机可以在任意地点起飞，无需特殊的地面配套设施，诸如飞行跑道、航空燃油、地勤人员等的辅助，非常方便普通科研工作者进行航空影像的采集。 而同时，随着计算机视觉技术的不断发展，尤其是多视图立体匹配与三维重建技术的进步，人们能够通过无人机拍摄的高分辨率航空影像，利用多视图场景三维重建算法恢复出航拍区域的三维模型。 在可以预见的将来，伴随着国家低空空域的进一步放开，通过无人机拍摄的航空影像数据会呈现巨大乃至爆炸性的增长，这些数据蕴含着丰富的潜在价值，因此对无人机航空影像的应用进行研究，无论从学术上还是商业上考量，都具有重要的意义。 本文针对基于无人机影像的正射全景图像生成和三维场景纹理映射进行了深入的研究。通过无人机航空影像，生成航拍区域的正射全景图像，并对重建得到的三维场景进行纹理映射，得到具备真实纹理表面的三维模型。 本文的主要工作包括以下几个方面： 第一，本文基于无人机航空影像实现了三维场景的纹理映射。本文基于无人机航空影像，利用多视图场景三维重建技术生成航拍区域的三维点云模型，进一步对点云模型所表述的三维场景进行纹理映射，得到具备真实纹理表面的三维模型。与传统的数字高程模型对比，本文的三维模型不但能够对地表高程起伏变化进行描述，同时包含有地表的纹理信息。而相较于三维点云模型，本文的三维模型不存在孔洞或缺损，在小尺度下的表现效果更优，细节表现能力更强； 第二，本文提出了一种针对无人机航空影像的正射全景图像生成算法。无人机飞行高度低，其航空影像的空间分辨率相较传统飞行平台采集得到的航空影像更高。但是由于无人机质量较小，飞行姿态易受气流扰动，其航空影像容易出现较大倾角，同时较低的飞行高度也使得复杂的地貌在成像过程中容易发生明显的畸变。相较于现有的图像拼接算法，本文提出的方法可以有效地应对倾角过大、地表高程变化所致的无人机航空影像畸变问题，所生成的航拍区域全景图像具备良好的正射性能，且不含有明显的缝隙、畸变、扭曲或者拼接错误； 第三，本文的方法具有较强的适用性，可以有效地处理包含平原、丘陵、高地、山地、河流、村落等不同地形地貌特征的无人机航空影像。同时该方法能够满足大规模数据处理的需求，在我们的实验中，最大的无人机数据集包含有1125 幅航空影像，覆盖航拍区域面积最大达到122 平方公里。三维模型和正射全景图像均...; The 3D model of earth’s surface and aerial panorama play an active role in various fields, such as transportation onstruction, hydraulic engineering, ecological research, urban planning and national security. For a long time, esearchers have done deep research on 3D reconstruction of earth’s surface and aerial panorama generation. raditionally, people apply LiDAR or SAR carried by fixed-wing planes or satellites to obtain the digital elevation model (DEM) for the use of 3D model of earth’s surface. While the aerial panorama is made from airborne images collected by sky-based platforms. In recent years, UAV (Unmanned Aerial Vehicle) technology spread from military fields to civil applications, UAV has become a novel and popular means to collect airborne images. Compared with fixed-wing planes or satellites, UAV has salient advantages, such as smaller size, lighter weight and the convenience to be carried with. Rotary-wing UAV can take off at any location without special support from ground, e.g. runways, aviation fuel or ground crew, which is very convenient for common researchers to collect airborne images. Meanwhile, as computer vision technology develops, specially in fields of multi-view stereo and 3D reconstruction, people could restore the 3D model of the aerial area by using multi-view scene 3D reconstruction method from high resolution airborne images captured by UAV. In the predictable future, along with the further opening up of low-altitude airspace to civil uses, the amount of airborne images collected by UAV will produce a huge or even explosive growth. These data has rich potential value, so it is of great significance to work on UAV image application whether from academic or commercial considerations. This paper has focused on the ortho-rectified panorama generation and 3D scene texture mapping from UAV images. Based on UAV images, we generate the ortho-rectified panorama of the aerial area, meanwhile we do texture mapping on the reconstructed 3D scene to obtain the 3D model with realistic texture surface. Our main work can be summarised as: 1. We accomplish the texture mapping on the 3D scene based on UAV images. In our work we create the 3D point cloud of the aerial area from UAV images by using multi-view scene 3D reconstruction method. Further we do the texture mapping on the 3D scene represented by the 3D point cloud to obtain the 3D model with realistic texture surface. Our 3D model is superior over the traditional digital ...
修改评论