Analysis of 3D Model Construction and Accuracy Assessment by Using Multi-View Images.

• Main Authors: Chen, Chien-Hong, 陳建宏
• Other Authors: Hwang, Jin-Tsong
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/29914012811976507413

碩士 === 國立臺北大學 === 不動產與城鄉環境學系 === 103 === This study will present an approach to achieve the goal of 3D model reconstruction without expensive equipment and less human intervention. This study tries to rebuild three-dimensional models by the method which applies the principle of photogrammetry that feature points extraction is different from close-range photogrammetry. VisualSfM is a free software which proceed with image-based modeling through the reconstruction of spatial geometry by searching for feature points to match and stitch images, restore the camera positions, inquire the coordinates of shooting subject, and navigation systems to provide users with browsing their own image location. The spatial geometry reconstruction process is divided into three steps: feature points extraction, feature points matching, and restoring the camera position. The first two steps are based on SIFT (Scale-Invariant Feature Transform) algorithm, the last step is based on SfM (Structure from Motion). This study would be used a consumer digital camera, a general smartphone, and a digital single-lens reflex camera to rebuild the experimental building with pre-laid targets, and applies the difference of a known target coordinates in RMSE as the accuracy evaluation. The result shows that the RMSE range of these three cameras are between 1.1 cm to 2.7 cm. The digital single-lens reflex camera is the best, the consumer digital camera followed, and the general smartphone is the worst. This approach is not only able to record the space information conveniently and reliably, but also provides another option for the future construction of a three-dimensional model which achieve the LOD2 (Level of Detail 2) level.