Feature Extraction from Dense Image Matching Point Clouds of Buildings by Using Tensor Voting Method and Voxel Method
碩士 === 國立成功大學 === 測量及空間資訊學系 === 104 === In this thesis, Tensor Voting Method(TVM) and Voxel Method(VM) are used to extract the geometry features, such as points, lines, planes and volume, from the dense matching point clouds of buildings. Moreover, VM could be used to detect blunders in the dense po...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
2016
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/88u9y5 |
| Summary: | 碩士 === 國立成功大學 === 測量及空間資訊學系 === 104 === In this thesis, Tensor Voting Method(TVM) and Voxel Method(VM) are used to extract the geometry features, such as points, lines, planes and volume, from the dense matching point clouds of buildings. Moreover, VM could be used to detect blunders in the dense point clouds. After the simulated data are used to analyze the factors that could affect on the result of these two methods, dense image matching point clouds of three buildings by using 30 aerial images are used to derive their geometry features. The extraction result of simulated data indicates that TVM is likely to fail to extract features possibly due to low signal-to-noise ratio of points on local planes of buildings. On the other hand, VM is tested to analyze how the density and distribution of dense points on building surface could affect feature extraction quality. The result of Voxel Method with the use of real data shows the RMSD of point coordinates determined is about 1.21~2.32 GSD by comparing with the corresponding ground check data. The edges determined by Voxel method and the check data have the distances of about 1.03~5.05 GSD and RMSD of about 0.05~1.11 GSD.
|
|---|