Automatic 3D coordinate estimation of feature points for building modeling using stereo images

In dense urban areas, close-range digital photogrammetry is appropriate for three-dimensional (3D) modeling in terms of cost and portability. In this method, passpoints between stereo images must be extracted in the orientation process. Because of the number of buildings and obstacles in urban areas, automatic selection of passpoints is complex, whereas manual selection is time-consuming. However, for detailed modeling, components of a building must be extracted. Therefore, the method is developed to automatically extract components of a building for efficient modeling in this study. In the proposed method, homogenous surfaces are extracted by using RGB bright values and passpoints are accurately matched by using an association program automatically. Then, 3D coordinates are calculated. This method was validated by using photographic images taken near Kodaiji temple. The RMSE was about 16 cm, showing that the proposed method of automatic estimation of feature points generates acceptable models in terms of accuracy.