We address the problem of efficient image matching for large, highly redundant photo collections with highly complex topologies, such as photos acquired by unmanned aerial vehicles (UAV), focusing on disaster monitoring. Our approach conducts a skeleton graph which simplifies the image topology with the consideration of image importance and topological relationship. We define the image with the highest importance weight as candidate, adding the remaining images referring to the topology skeleton. To conduct the skeletal graph, the image topology is first computed depending on the overlapping relationships between images. Experimental results show that our technique drastically limits the searching range that is for feature similarity computation, resulting in dramatic speed up. A final bundler adjustment is implemented in the procedure of scene reconstruction, and the completeness and accuracy are far more comparable to the traditional method.