Feature Matching Research Of The Autonomous Aerial Refueling Based On Binocular Vision

On the background of aerial refueling, this paper systematically analyzed binocular vision stereo matching algorithm of the image based on computer vision measurement technology. The binocular stereo vision is photographing the same object from different angles, and calculated the image parallax by the geometric imaging principle position deviation, and calculation three-dimensional informationIn this paper, we study the binocular stereo vision technology in the image acquisition:camera calibration, feature point extraction and stereo matching. Firstly, this paper introduces the research background of aerial refueling meaning and machine vision research status at home and abroad, and then to discuss and research how to realize the various technical problems in this research topic. Secondly, Camera imaging was introduced in detail and because the basic model information requirements good stability and high precision for unmanned aerial vehicle in aerial refueling, while the traditional calibration precision is not good enough, so in this paper, by using the stereo vision system based on polar correction calibration algorithm to get a more accurate result. After obtain the inter parameter and the external parameter, adopt epipolar line rectification calibration, and considering the influence of tangential distortion, to make the calibration result more ideal and accuracy. Thirdly, in view of the traditional SIFT (Scale Invariant Feature Transform) descriptor in image processing, there are multiple similar areas likely to cause the mistaken match while the Daisy can match efficiency that can be reduced by descriptors, adopt stereo matching algorithm based on SIFT and Daisy description operator, the method using the SIFT algorithm to generate the key Feature points, reuse the good rotation invariance of Daisy descriptor itself, describing feature points by using Daisy descriptor instead of original SIFT Feature description algorithm, then adopts the Euclidean distance of Daisy character description operator as a judge to match similarity measure criteria, and then use the RANSAC to eliminate the false matching points, and good matching image effect is obtained. The simulation results show that it has good robustness on image match and satisfy the requirement of aerial refueling docking in real-time and high accuracy.