|其他摘要||The study on 3D reconstruction technique is a hot subject and a variety of 3D city modelling methods have been increasingly applied in a growing number of fields. Many early used 3D modelling softwares, such as early used AutoCAD, 3DMax, accomplish 3D reconstruction according to the real scene. These methods can only show low levels of detail contour features of urban landscape. Using actual image textures and point cloud data obtained by aero-photogrammetry system and airborne radar system to accomplish 3D reconstruction is rising. Because the efficiency of using sensors (i.e. photographic camera and laser scanner) carrying on unmanned aerial vehicle to obtain images is very high and different precisions can be achieved based on different resolution images, the images acquired by unmanned aerial vehicle technology are widely used in large-scale 3D city building reconstruction. Close-range-photogrammetry is another important methods that is used in 3D reconstruction technique. Compared to images obtained by unmanned aerial vehicle technology, close-range photogrammetry can improve the precisions of 3D modelling based on following two characteristics: first, because close-range photogrammetry can obtain ground images at a much closer distance, the ground images have higher resolutions and more information; second, because the inclined angle of camera axis between ground images obtained by close-range photogrammetry and images acquired by unmanned aerial vehicle exceeds 90°, which makes the angle information more abundant. To establish 3D model with higher precision, the thesis focuses on the modeling methods of ground close range images. The main novel contributions in this project are described in the following:
(1) Camera calibration and precision analysis were finished. This work came up with some methods for camera calibration, and the calibration results were analyzed. We calibrated variable parameters of our camera with RMSE of 0.0018mm. Two methods were used to validate the stability of calibrations and results shown that the camera calibration is reliable.
(2) The design of photography plan was proper. This work made several aerial and ground photography plans according to the conditions of the observation regions and the requirements of mapping. Aerial images capture method mainly includes the flight area determination, course design, overlapping degrees determine, etc. Ground close shot images acquisition methods are described as: parallel multi-baseline integrity more photography, parallel multi-baseline high-density integrity photography, rotating multi-baseline photography. Furthermore, a photography solution was presented based on the vertical direction of the rotor aircraft.
(3) Validation and evaluation of ground images photographic approaches. In order to verify the feasibility of ground images photographic approaches, the ground images were modelled and then the absolute orientation and relative orientation were combined to validate and evaluate the precision of proposed models The validation results shown that the precision of all photographic approaches below 8cm, and normal case photography based on reference points is the best photographic approach
(4) Calculation experiment for the combination of aerial images and ground images was finished. Image matching and point cloud matching were implemented in this thesis. The image matching program was proposed based on the SURF - RANSAC algorithm and then the matching experiments were conducted according to the proposed program. The results shown that the best matching effect is obtained by the same source data. And then we took several experiments to match the point cloud data of the two sorts of data, and finally we got the nested result.|
池梦群. 地面影像三维重建技术研究[D]. 北京. 中国科学院大学,2015.