Research Of Vision-based UAV Automatic Landing Correlative Technology

Unmanned Aerial Vehicle(UAV) have broad application prospects inmilitary and civilian. Taking back safely and reliably after given taskscompleted is the basic requirement for UAV, which is for re-using.However, the research and application of the recycling of UAV system stillhave many problems in currently. Research data show that the failure ofrecycling process causing flight accidents in the total flight accidents ismore than80percent. In this paper, vision-based UAV automatic landingtechnology is studied detailedly, and the study include the following threeaspects:(1) The detection methods of moving target in the case of cameramoving. Firstly, a moving target detection method based on motioncompensation is studied. The method changes the dynamic background tostatic by compensating the global dynamic background. Because themethod involves corner detection, matching and motion parametersestimation, the detection process are complicated, and have large computation. For these problems, the thesis presents a moving targetdetection method based on YCbCr color space in the case of dynamicbackground. The method establishes the color histogram firstly by thecolor model of YCbCr, then do image segmentation, then detect themoving target after some subsequent processing. Experimental resultsshow that this method is superior to commonly moving target detectionmethod based on motion compensation, and better real-time.(2) Vision-based pose estimation methods for UAV. According to thecamera projection principle, two pose estimation methods for theexperimental platform are presented, which is based on monocular visionand dynamic stereo vision-based UAV pose estimation method. Then a setof indoor UAV position and attitude estimation simulation system isdesigned, and the strong practical LED light source is used as cooperationtarget. Simulation results show that the estimation accuracy of thedynamic stereo vision-based pose estimation method is better than basedon monocular vision.(3) Vision-based moving target tracking methods. In order to lettingUAV landing safely and reliably, first it need vision algorithms to track themoving platform. For vision tracking problem, UKF-based moving target tracking method is studied firstly, and then the moving target trackingmethod based on wave gate tracking method is studied. The experimentalresults show that UKF-based moving target tracking method can predictthe trajectory of the tracking target, but its operation is very large andcumbersome process. The moving target tracking method based on wavegate tracking method can well predict the trajectory of the tracking targetand track it well. Because using wave gate tracking theory, the trackingalgorithm have good real time and good anti-noise. So, the trackingalgorithm has very practical value in the tracking of moving target.