The Design Of Vision-based Position And Attitude Estimator For Unmanned Aerial Vehicles Autonomous Landing

In this paper,a pose estimator based on embedded platform is studied and designed to solve the problem of pose estimation based on visual autonomous landing engineering application for medium or large fixed wing UAV.This paper first discusses the general scheme of pose estimator.Based on the visual autonomous landing demand,the components of the pose estimator are analyzed and the functions and requirements of each part are discussed.Then,the parameter requirments of the pose estimator components are obtained by analyzing the vision-based UAV landing process.According to the above analysis,the estimator is composed of a lens,a camera and a visual processing platform based on SoC FPGA.According to the overall scheme,the development platform of pose estimator is established to verify the hardware system and visual algorithm of pose estimator.Firstly,based on the Qsys model,the hardware of the pose estimator is built on the development platform.Then,according to the requirement of vision algorithm,the Linux system with OpenCV is set up on the development platform and the method of reading the camera data is put forward.Finally,in the development platform a hardware accelerating system is designed to accelerate the visual algorithm using FPGA to solve the problem of image data flow between ARM and FPGA.Then,the paper discusses the method and the way of implementing a series of algorithms of pose estimation on the development platform.Firstly,the method of position and attitude parameter estimation using algorithms including runway detection,runway feature extraction,runway recognition,runway tracking and position and attitude is discussed.Secondly,the implementation of the algorithm described above is described,and the method of modularizing the algorithm based on C ++ is discussed.The function of the position and attitude estimator is verified by a series of experiments.Firstly,the software and hardware system of the estimator is tested on the development platform.Secondly,the function of camera video reading and visual algorithm accelerating system are verified.Then the algorithm of position and attitude estimation is verified.Finally,the ability of the position and attitude estimator to output pose parameters is verified.Experiments show that the proposed position and attitude estimator has the ability to capture the landing image to the output pose parameter.