Design And Implemention Of Scene Matching System Based On UAV Image

Along with the rapid development of modern science and technology, the development of modern war tends to be automotive and unmanned. The UAV technology has become the key factor in modern warfare. At present, countries all over the world have plunged much manpower and resources to study the application of UAV. Most applications of UAV need to analysis the real-time image data and are based on scene matching. Therefore, it is very important to design a scene matching system for UAV.To match the real-time image of UAV with satellite map, this thesis designs a real-time scene matching system. First of all, track the corners of real-time image using optical flow method; propose a rule to judge track validity based on corners’distance stability. Estimate the movement of background through matching relationship of adjacent frames’ corners. Compensate the video’s motion in order to eliminate the influence on scene matching. Second, extract both the real-time image and its reference image’s contours. The contours of these images are matched by comparing their Hu invariable moments. This feature is very suitable for scene matching of UAV because the contour’s Hu moment is invariable when its direction, scale or position is different. This thesis designs a rule to filter the contours, which decreases the number of contours needed. Therefore, system’s matching efficiency is enhanced. Third, this thesis presents a method to estimate the transform relations between a real-time image and a reference image, using matched contours. It mainly makes use of the contours’ barycenter, enclosing rectangle and the rectangle’s center point to calculate the angle, scale and offset between the two images. This method is easy and can meet real-time requirement of UAV. Finally, judge the matching of the two images; if not the system uses the Kalman filter to predict the transform between them.This thesis firstly introduces current situations of scene matching; secondly, it describes some classical algorithms which are used in this system; thirdly, the system design and function models are proposed; it illustrates the improvement of the stabilization and matching modules in detail; finally, experiments are carried out with OpenCV library. The results demonstrate the system has higher matching rate, robust and could provide real-time property.