针对增量式运动恢复结构算法在无人机影像三维重建中运行效率低的问题，提出了一种基于关系图的无人机影像批处理重建方法。利用无人机低精度的位置姿态测量系统数据概略估算影像间的重叠关系，在关联影像之间进行尺度不变特征变换匹配，使用随机抽取一致性算法和三视图约束进一步剔除弱关联的影像连接关系，再通过深度优先搜索得到最终参加重建的影像集连通分量，构建出强关联的稳健影像关系图。然后，根据影像相对位姿参数，由最小二乘法解算全局一致性旋转参数，利用三视几何约束和线性规划求解全局一致位置参数，最后进行一次光束法平差优化。实验结果表明: 与经典的增量式重建Bundler算法相比，该方法运行速度至少提高了2.6倍，且生成的三维点数量增加了76.5%，更真实地恢复出场景的几何形态。该算法在改善重建效果的同时提高了处理效率，能够满足无人机快速响应应用的需求。

In response to low operation efficiency of the incremental algorithm for structure recovery from motion in three-dimensional reconstruction of UAV (Unmanned Aerial Vehicle) image, a diagram-based method for reconstruction of UAV image was proposed. In this method, data obtained from low-precision position and orientation system of UAV was adopted to roughly estimate the overlapping relationship between images, SIFT matching was conducted between related images, random sample consensus algorithm and three views constraint were used to weed out the connected relation between weakly-related images, depth-first search was applied to obtain the connected component of the images finally involved in reconstruction, and then a steady diagram for strongly-related images was designed. Then based on the relative position and attitude parameter of the images, the least square method was adopted to work out the consistency rotation parameter, the three views constraint and linear programming were used to solve the consistency location parameter, and then a bundle method-based adjustment optimization was conducted. It indicates that the operating speed of this method is at least 2.6 times faster than that of the typical incremental Bundler algorithm, the three-dimensional points increased by 76.5%, and it can more closely recover the original geometry form. This method, while perfecting the reconstruction effect, improves the treatment efficiency and can meet the requirement for quick response of UAV.