Geographic Information Labeling On Aerial Video

Because of the dramatic increase of aerial images and videos caused by the development of aviation area and shoot techniques becoming more and more easy, aerial images and videos have been applied in many fields. How to research based on these aerial images and videos becomes a hot issue among the researchers in various fields. But the process of aerial images and videos requires a lot of work and much other relevant auxiliary information, which limits their application in various fields. It is useful to improve working efficiency for analysts of aerial videos to analyse aerial images and videos with image processing methods.In this paper, the study object is aerial videos of UAV videos, doing research on labeling geographic information on aerial video, and the study focuses on the track of adjacent frames, road extraction of aerial images and the matching of aerial and orthophoto. The design and implementation of labeling geographic information on aerial videos provide technical support for relevant researchers to attain intelligence information quickly. The main work of the thesis is as follows:1. A method of road extraction of aerial image based on clustering segmentation and Hough transform is proposed. The road extraction is interfered by cars, buildings and their shadows, which lead to some roads lost or wrong. With the road features of aerial images, the road network is extracted by K-means clustering and Hough transform. This method could resist these interference to some extent.2. A method of geographic information acquisition of aerial image based on orthophoto is proposed. With the consistency between aerial image and geographic, the road extraction is applied. With the approach of line feature matching based on eigenvectors, the correspondence between aerial image and orthophoto is determined, by which, the geographic information of aerial image is determined.3. A good global motion tracking method of video is identified. By the study features of video analysis and comparison of the performances of several different methods of point feature extraction, the SURF algorithm is determined to match point features. Then, RANSAC is applied to make the global motion tracking become true combined with the strategy of matching a long sequence of video frames. The experiment verifies that the precision and operational efficiency is relatively high.4. Geographic information labeling on aerial video based on orthophoto is introduced. With the geographic information of the first video frame and video tracking, the labeling of geographic information on subsequent video frames is achieved, which verifies the validity of above methods and provides a basis for video analysis.