|其他摘要||In recent years, as the demand for three-dimensional model of the city, the traditional down-looking lens, can obtain height information and texture information at the top of the main building, but the nadir shot from aerial fly is inadequate to acquire enough side texture information for 3D modeling.
Compared with traditional aerial photography, oblique aerial photography can collecting ground images from multiple angles at the same time, thereby overcoming the limitations of traditional aerial photography that can only be taken from a vertical angle. Oblique aerial photography can reflect the surface features in more high-fidelity. Currently, oblique photography technology based 3d city modeling and reconstruction is a development trend of three-dimensional city modeling.
For oblique aerial images, this paper describes a method to reconstruct three-dimensional scene based on computer vision. The main technical part is made up of feature extraction, image posture calibration, multi-angle imaging dense matching, etc. The specific tasks are described in the following:
(1) The oblique aerial image has the characteristics of large dip Angle and a high degree of overlapping regions. The perspective of the traditional aerial triangulation for aerial image needs more stringent requirements, and it is not suitable for aerial triangulation for oblique aerial images. We improve the process of aerial triangulation via computer vision approach in the first step. Then we extract the image feature points, spatial relationships between image and calculating the exterior orientation elements.
(2) We get the result of aerial triangulation using oblique aerial images. The combination of different lenses and the different overlapping degree of images, which have a direct impact for the accuracy of image dense matching. Our experiments show that the modeling results of five lenses of aerial triangulation is the best, and the accuracy of using four lens is weak due to the lower overlapping degree. Furthermore, we have done the systematic analysis of the accuracy of aerial triangulation as well as the sources of several errors which affect the result of the experiment, such as flight quality, whether the control point selection is appropriate, the improper operation, and so on.
(3) The paper introduces the method of Delaunay triangulation for point clouds. Compared with several methods of point cloud subdivision, elaborates the method of texture mapping.
(4) We analyze the accuracy of three-dimensional model as well as the sources of several errors which affect the result of the experiments. According to the accuracy evaluation of reconstruction of 3D model in plane accuracy, elevation accuracy and performance accuracy. An fact can be found that the result of reconstruction of 3D model meet standard model requirements.|
马晨. 航空影像三维重建技术研究[D]. 北京. 中国科学院大学,2015.