Low-altitude unmanned aerial vehicle (UAV) images shave increasingly demonstrated unique advantages in the fields of transport planning, urban management, emergency monitoring. To understand the advantages of UAV images, rapid and accurate correction and registration of the image are a necessary premise. This paper takes a image of intersection taken by hovering UAV as its research target, and discusses the method of geometric correction and registration for such images. The image contains a large number of similar features and moving objects, which makes it difficult to be corrected in traditional way, based on spatial domain or frequency domain. Thus, this paper proposes an image registering method based on semantic information extracted from zebra lines. To realize images registration between arbitrary frames, the following steps are necessary: patterns extraction, zebra line patterns preparation and classification, axis fitting and image registration. Finally, a test is performed with a video taken by UAV. The results demonstrate that the method is robust and effective.