Research On Vision-based For Landmark Tarcking And Landing Of Unmanned Helicopter

Because of its low cost, high performance and easy maintenance etc, the unmannedhelicopter with autonomous flight ability is widely used in military and civil aspects. Theunmanned helicopter with autonomous flight ability is also of a very high scientific researchvalue. Binocular stereo vision is an important branch of computer vision research, which canget the depth information of scene flexibly and accurately in a variety of conditions. Therefore,it is of great significance for the flight stability and safe landing of unmanned helicopter in theunmanned helicopter landing control based on vision to get image feature by using binocularvision and to get the relative pose of unmanned helicopter and parking apron by usingposition attitude estimation algorithm. This paper completed the following work:First of all, I have designed a landing platform design in this paper according to apronsused in various experiment which down-regulated the middle of l the traditional "H" shapedlandmark making the size of two parts inconsistent. This paper aims to use the improved "H"shaped landmark to simulate the apron.Secondly, aiming at the landing platform fixed in the airport, this paper puts forward areal-time algorithm which can identify and estimate the unmanned helicopter own poseinformation. First to pretreat the image obtained by the two marked cameras. Then, segmentthe landmark information through the image analysis and processing. To improve theprecision of the characteristic image processing will greatly improve the accuracy of poseestimation. This paper, by using the sub-pixel level Harris operator to collect the angle pointof the two images and to put all the characteristic angle point mark out and to match. Andthen estimate the relative unmanned helicopter landing platform6bits appearance variable.The experimental results show that image processing and pose estimation accuracy andcomputation speeds meet the real-time requirements, the result of calculation is reliable, theerror is in the permitted range.Aiming at the situation of landing platform moves, this paper proposes a trackingalgorithm based on Kalman filtering technology combined with the binocular vision system.By using of landmark’s characteristic, combining profile fitting realize landmark recognition,to get landmark model and cover the landmark with rectangular window; Using Kalman filter to estimate each frame of the initial iteration position and using binocular vision system fortracking position, then to feed back it to Kalman filter. This tracking algorithm can predictand estimate the condition of mobile landmark, so it still can be on the target track accuratelyin the short time of the loss of landmark. The experimental results show that the algorithm canrealize the detection and continuous tracking of mobile landmark and has good robustness forkeeping out as well.