Multiple images matching for 3D building modeling

• Main Authors: Yu-Hsiang Huang, 黃祐祥
• Other Authors: Liang-Chien Chen
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/61274739451534849319

碩士 === 國立中央大學 === 土木工程研究所 === 98 === Three dimensional geographic information systems describe the reality with environmental elements such as buildings, roads, vegetation etc. Among those elements, building is the most prominent. Aerial images are commonly used in building reconstruction. In building reconstruction with aerial images, building detection, image matching, and building modeling are the three major works. This study uses multiple strips acquired from highly overlapped aerial images to perform 3D building modeling. Multiple image matching and building modeling are the major works in this study. Test cases include different types of building roofs. We first select the working area in the master image to extract building features. Then, the detection of building information with line features can aid the image matching. Multiple image matching employs multi-windows with multi-epipolar constraint to find conjugate positions in the same procedure. In the generation of line segments, line tracing is performed in the image space with 3D information. Building modeling is made by the integration building boundaries and roof surface information. The results show that multiple images matching can improve the matching accuracy. Model refinement with 3D roof point clouds can work well. Compared with reference data, the RMSE of building model in most of case can be smaller than 0.4m.