Research And Implementation Of Vision-based Autonomous Landing Of UAV

In recent years,quadrotor UAVs have been widely used in everyday life due to their simple structure,controllability and vertical take-off and landing capabilities.Whereas autonomous landing is their key technology,although some progress has recently been made in GPS-assisted navigation of quadcopters,concerns about collisions still overshadow their reliability and safety,especially in environments where GPS is not available.As a result,autonomous and safe landings that rely on visual positioning have become a research priority.In this paper,an autonomous landing system is designed using a quadrotor UAV as a vehicle to estimate the position of the UAV through monocular vision.The main research elements of this paper are:Firstly,a landing marker consisting of multiple nested Ar Uco markers is designed to solve the problem of easy loss of moving targets.Further,the feature points obtained from the target identification process are used to construct the Pn P equation and solve the relative positional relationship between the marker and the camera.To solve the problem of tracking accuracy of moving targets,an interactive multi-model algorithm based on the traceless Kalman filter is used to achieve the estimation of the real state of the target.Tracking simulation experiments are carried out for uniform linear and uniform turning motion models,and the results show that the algorithm achieves good results.Secondly,a BP neural network PID controller based on particle swarm algorithm optimization is adopted to address the disadvantages of traditional PID control algorithm such as poor adaptivity and parameters cannot be adjusted in real time.Based on the adaptive control of BP neural network PID,the initial weights of the neural network are optimised using the global optimality and rapidity of the particle swarm algorithm,which solves the problems of BP neural network easily falling into local minima and slow convergence speed.The results show that it has faster response speed and higher accuracy as well as good robustness compared with BP neural network PID and traditional PID control methods.In addition,a quadrotor fixed-point hovering simulation experiment is carried out and the results show that a stable hovering effect is achieved.Finally,the overall framework of the autonomous landing system was designed,and the hardware platform selection and the experimental platform construction were determined.Simulation experiments of stationary target landing and moving target tracking landing were carried out sequentially under ROS and Gazebo environments to verify the stability of the landing system.Then the actual UAV autonomous flight test was carried out to verify the reliability of the autonomous landing system in this paper.