Practical modeling technique for large-scale 3D building models from ground images

In this paper, we propose a practical modeling method for large-scale building modeling with multiple image sequences. First, a scene is acquired via a pose camera, which annotates each image with navigation information using GPS and a digital compass. An adjacency graph was built to determine the approximate spatial adjacency given roughly known poses of each image sequences, to organize the nodes into a set of sequences and to access the nodes. The camera positions are obtained by using a standard Structure from Motion algorithm. After the pose estimation, the user-assistant modeling approach is used based on a half-edge data structure. For accuracy and validity in the approach, some experimental results are shown.