Comparing with manned aircraft, an Unmanned Aerial Vehicle (UAV) is a more flexible with lower cost platform for aerial photo acquisition. However, its payload, space, and endurance time is comparably lower and generally only light and small format consumer grade digital camera can be carried out. That means, its footprint coverage will be smaller and more images are necessary to cover the same area, and may contain only forest within one image when flying over the mountainous area. Thus, it is a difficult to utilize the conventional aerial triangulation (AT) procedure to obtain the images' exterior orientation parameters (EOPs) that requires uniform distributed tie-points within the images. Because, it is difficult to match tie-points automatically due to the image context has repetitive pattern, shadow, and homogeneous area. Since the geological hazard happened frequently after heavy rainfall in Taiwan mountainous area, it is thus suggested to utilize a fixed-wing UAV equipped with a tactical grade SPAN-CPT IMU together with a dual-frequency GPS antenna, and a Canon EOS 5D Mark II digital camera for direct georeferencing (DG) particularly when fast response for hazard investigation is required. In this paper, an in-door camera calibration field is designed for the calibration of interior orientation parameters (IOPs) and an outdoor calibration field with two-step boresight calibration procedure is applied for the purpose of DG. Detail about the system calibration procedure and accuracy analyses will be provided in the paper. Experimental results show that for flying height with 1200 m after DG and performing accuracy assessment using space intersection through check points, the RMSE in planimetric and vertical directions are less than 1 m and 4 m, respectively. It demonstrates that the positing accuracy after DG is enough for the purpose of fast hazard area investigation. On the test of topographic mapping, the internal precision of the generated DSM through ground controlled and non-ground controlled aerial triangulation are all about 1 m, while the planimetric accuracy in stereo-mapping is within 0.5 m. If two UAV image datasets can be acquired before and after disaster, the landslide loss and deposit volume could also be estimated.