Aiming at the difficulties of image matching and aerotriangulation for unmanned aerial vehical image in the area with high altitude drop, this paper presents a stepwise refinement method for image matching and aerotriangulation using correction of local relative geometric distortion (CLRGD) which is based on the initial aerotriangulation and rough terrain model. It takes into account the image relative distortion caused by relative tilt of stereo image pairs and topographic relief in which an algorithm for single point accurate matching considering CLRGD has been adopted in the process of normal aerotriangulation. Then image matching and orientation can complement each other so as to improve the accuracy and reliability of matching and aerotriangulation. Experimental analysis proves that the algorithm can do well in the image relative geometric distortion, strengthen the regional network and further improve the accuracy of the aerotriangulation in the area with high altitude drop. In terms of efficiency, the image range of CLRGD is relatively limited due to the use of rough terrain model, so the algorithm can basically meets the actual production needs. If considering that the accurate single point matching used by each tie point has good independence, a parallel algorithm can further improve the efficiency.