Vision Based Autonomous Landing Navigation And Implement For UAV

With the development of unmanned aerial vehicle(UAV),the requirement of autonomous flight for UAVs is higher and higher.The autonomous landing is the most complex and challenging task in UAV autonomous flight.On the one hand,autonomous landing requires high-precision pose estimation and control,on the other hand,most conventional methods of navigation such as inertial navigation,GPS and so on are used in landing navigation at present.The environment is limited and susceptible to interference,and the accuracy is poor.With the development of computer vision technology,visual information can be used to assist navigation during landing.Visual navigation can provide relative position and attitude parameters,high accuracy,anti electronic interference,and can identify obstacles and improve landing safety.Based on the above background,this paper is going to study the vision based UAV autonomous landing navigation algorithm and hardware platform implementation.The main contents are as follows:This paper analyzes the research status of UAV navigation,and the processing platform firstly.In order to realize the real-time acquisition and processing of the landmark image,based on the idea of modular design,by the means of multi-processor of “ARM Acquisition+DSP Processing”,this paper will build a visual navigation system hardware platform.It mainly includes the design of visual image acquisition terminal,processing terminal and network communication between the acquisition end and the processing unit.The acquisition and processing of airborne real-time online are realized.In order to realize the autonomous landing of UAV,a landing sign with special shape is designed in this paper.Through the airborne sensor to collect landmark images,considering the noise in the process of collecting and other redundant information,the first image sequence was related to pretreatment.Using ORB algorithm to extract the feature points in the image are landmarks,location and description information.On this basis,the corresponding relation of feature points in two frames is obtained by matching algorithm.For the problem of vision pose estimation,firstly,the coordinate system is defined,and the vision pose estimation model is given.With the inter frame motion estimation method,the homography matrix is decomposed by the correspondence of the feature points among the frames,and the relative pose of the UAV is determined by coordinate transformation.To solve the mismatching problem of feature points,the random sampling consensus algorithm is adopted to optimize the homography matrix.Aiming at the realization of vision navigation algorithm in multi-core DSP,the parallel architecture of multi-core DSP is studied.The multi kernel application of image edge extraction is designed by using IPC master-slave mode and OpenMP respectively.The experimental results show that the application of multi-core architecture can greatly improve the real-time and processing capability of the system.Based on the C6678 platform,the algorithm is implemented on a single kernel,and the correctness of the algorithm is verified.On this basis,the multi-core application of visual navigation algorithm is designed to realize multi-core parallel processing and improve the real-time processing ability of the system.