Abstract: Aiming at the problem that traditional photogrammetry theory has high requirements for aerial photography conditions such as route planning, flight attitude and image overlap, yet unmanned aerial vehicle (UAV) has no obvious advantage of efficient support. Based on the theory of computer vision, a robust structure from motion(SFM) technique is proposed in this paper. Firstly, the matrix expression of the spatial rotation relation is transformed into the linear expression of the vector by using the rotation average method of Lie algebra. Then, before the least square adjustment, the L₁ norm is introduced to optimize the initial value iteratively to solve the global consistent rotation parameters. Finally, the coordinate system of displacement and rotation parameters are unified to realize the 3D coordinate calculation of matching points. The experimental results show that, the UAV image 3D reconstruction technology based on global SFM has higher accuracy than the traditional photogrammetry method, and the 3D point cloud reconstruction effect is better. The accuracy of point position measurement is better than 0.3 m under the beam adjustment aided by differential GPS station coordinates. Under different route layout conditions, the success rate of image solution is 100%.

Key words: global SFM, 3D reconstruction, Lie algebra, least squares, ground positioning