Abstract: A low-cost, simple and efficient method of 3D reconstruction of video frame images based on structure from motion (SfM) is proposed for the problem of realistic 3D modeling of complex shaped ancient buildings. Firstly, a single-lens consumer-grade UAV is selected to obtain high-resolution video frame images by designing and planning a reasonable flight path, and generate detailed 3D models of complex shaped ancient buildings using SfM algorithms. Taking the Ming dynasty drum tower in Xingqing district, Yinchuan city, Ningxia as an example, we quantify the variability of the three platforms in terms of modeling efficiency and modeling quality, and explore the optimal platform using strategy. The results show that the 3D reconstruction method can efficiently achieve the acquisition of fine texture on the surface of complex heterogeneous ancient buildings and the reconstruction of realistic 3D models, in which the modeling effect of ContextCapture platform is better than PhotoScan and RealityCapture platform, and the modeling quality and texture clarity are better, so it is recommended to use ContextCapture platform for 3D reconstruction. The method in this paper can provide a basis for the 3D reconstruction strategy for the conservation of complex heterogeneous ancient buildings, which has the prospect of wide application.

Key words: structure from motion, 3D reconstruction, ancient buildings, digital conservation