Multi-View 3D reconstruction has been studied by researchers from various countries for decades, but still is a heat point in the field of computer vision. Generally 3D reconstruction system can be realized by the method of incremental Structure-from-Motion, which consists of several procedures such as feature detection, feature matching and triangulation. In this paper, we propose a speed-up 3D reconstruction system that associates several efficient algorithms, which can recover 3D structure form a set of unordered images of the ancient architectures in China. The experiment result shows that the system has a good performance in efficiency and final reconstruction appearance of kinds of buildings.