| Summary: | 碩士 === 國立政治大學 === 地政學系 === 106 === The source for facade texture-mapping for 3D building models can be obtained from aerial oblique images. However, the images of high building lower parts are difficult to be obtained in some areas. Close-range images can remedy such a shortcoming of aerial oblique images. Additionally, close-range images can be collected faster and more effieiently by Surveying Vechicle. However, before facade texture mapping, bundle adjustment should be implemented to orient these vehicle-based side photographic images. Generally speaking, tie points and control points as well as camera parameters are necessary in bundle adjustment. Since straight line features of, e.g. images of signboards, exist in these vehicle-based images, they can be also treated as tie lines in bundle adjustment. In this study, it is supposed that vertical distance from surveying vehicle to buildings is 3m, the overlap of adjacent images from one camera and two cameras is 60 percent and 80 percent, respectively, and the horizontal angle bwteen the optical axis between two cameras is about 100 degree. The purpose is to find out the most number and the best number of vehicle-based side photographed images in bundle adjustment on the basis of two set control points, totally four points, at both sides of test area and under the accuracy requirement of 3D LOD3 building model. The other purpose is to discuss the potential accuracy by using tie lines in bundle adjustment. The result shows that vehicle-based side photographic image orientation determination from simultaneously two cameras is better than that using images from one camera. The most number of vehicle-based side photographed images using simultaneously two cameras is 70, ca. 245m. The best number of using simultaneously two cameras is 28, ca. 90m. Although the other result shows that no accuracy improvement of vehicle-based side photographic image orientation determination with adding tie lines in bundle adjustment, the result still shows accuracy is little improved in orientation determination with adding tie lines in self-calibration bundle adjustment.
|
|---|
