Historical aerial photographs can directly witness the historical landscape environment and link to the spatial information in the past. Which are important information sources to evaluate the temporal spatial evolution of zones of interest. In Taiwan, abundant historical aerial images have been acquired and archived by Research Center for Humanities and Social Sciences (RCHSS) of Academia Sinica. However, camera report of images was usually confidential. Most of the historical aerial images haven't been registered since there was no precise POS system for orientation assisting in the past. To handle such great quantity of images, this study was processed through methods of computer vision. Image features were extracted and matched by Scale Invariant Feature Transform (SIFT). To improve matching accuracy and efficiency of conjugate points, an improved Random Sample Consensus (RANSAC) was promoted by considering the error and recording the largest areas covered by randomly sampling points. Then, according to adjacency matrix, the relationships of conjugate points were established and stored. After image matching and alignment automatically, the relative orientation of images can be acquired. For registration, control points are needed to be added to do network adjustment based on 2D coordinate transformation (2D affine and 2D projective). Finally, those feature points matched by this procedure can be used to build control image database, and all computation is based on point data instead of image data. Also, further study such as multi-temporal environmental changes can be investigated by using this temporal spatial data system.