Abstract: Image stitching is one of the most important core technologies in UAV research areas. In battlefield, the images captured by UAV will be stitched into a panorama picture in order to help commanding officers understand the battlefield dynamics and make strategic decisions. In this paper, a fast and accurate algorithm of stitching based on UAV images was proposed. Aerial images will be processed by feature detection and extraction, and then an innovative matching strategy within a predicted region, finally a feather blender. This algorithm is fast with good quality. The effectiveness and efficiency were confirmed using both Google Earth images and real UAV images in experiments.