Research On Key Techniques Of UAV’s Autonomous Landing Based On Vision

As one of advanced navigation systerms, vision-based navigation system has a lower cost, morestrong anti-interference ability than other traditional navigation systems. With UAV’s application anddevelopment in military and civilian fields，especially in the wars, related technologies about UAVhave been studied by many domestic and foreign universities and research institutions ororganizations in recent years.The surrounding environment perception and an accurate estimate of UAV position and orientationstate are two basic requirements of UAV visual navigation. During the process of UAV autonomouslanding, the landmark identification is one way to perceive its surrounding environment. Thedissertation aims at two big aspects, including the accurate detection of landmark and UAV’s poseestimation. The main work is as follows:(1)The background and content of the research are described in the dissertation. Then thetechnology state quo is presented on UAV’s landmark identification and pose estimation. A newlandmark, which is combined by6red circles with centers marked, is designed in this paper. Thelandmark can be easily separated from the surrounding environment, and meantime, it includesenough vision information for pose estimation. Apart from these, pose state can also be estimatedaccording to common tangent of two circles when part of the landmark is missing.(2)As for landmark identification, pattern recognition based on machine learning is used in thisdissertation considering time-consuming of traditional pattern-matching recognition. First of all, theimages taken by UAV during the flight can be affine transformated due to the loaded equipments orenvironment such as wind.So affine moment invariants are used as features of landmark identification.Second, it is difficult to get many real landmark images as training samples because of UAV flightenvironment’s unpredictability and differerent fight satus. As a result, SVM classifier is used as it cansolve the classification problems based on small samples. So in this paper, SVM classifer with affinemoment invariants is used to classify the landmark. Kernel and penalty parameter in SVM are chosenproperly.In the end, experiment results verify the effectiveness of this method.(3)In this paper, the motion and projection model of the camera on UAV is built. And a methodbased on designed landmark is proposed to estimate UAV pose state through the transformationrelation among image coordinator system, camera coordinate systerm and world coordinate system. Inthis method, based on the specifity of the landmark, centers of six circles are found by using harrisand at the meantime their coordinate in image coordinate system is recorded.Then, Levenberg-Marquardt algorithm is used to get UAV’s pose status such as pitch angle, yaw angle androll angle pose status.