Research On Vision Based Autonomous Landing Of Rotor UAV On A Moving Platform

The research of autonomous landing for rotor UAV is one of the most popular topics in the field of UAV research.Conventional UAV landing usually relies on inertial and GPS information.The application environments of conventional landing techniques are limited and susceptible to interference.Visual sensors can provide the UAV system more information about the environment,which improving the safety in UAV landing.In this thesis,we mainly focus on the vision guided autonomous landing of a rotorcraft on a moving landing platform.The main work is summarized as follows:Firstly,the autonomous landing system for rotor UAV is set up.A quadrotor is selected as the body of the rotor UAV system.The flight controller unit is designed using STM32F427 and FreeRTOS.The on-board Jetson TX1 and high-definition cameras are adopted as the visual unit.To address the problem of the limited field of view of a single camera,the guide mark detecting and tracking method is designed which combines the information from both the front-view and the down-view cameras.The visual guidance strategy is designed based on the method.Secondly,the detecting and tracking methods of the typical mark in the interference environment are designed and implemented for the two different on-board cameras,respectively.For the front-view camera,the moving platform and the guide mark are simultaneously detected using Tiny-YOLOv3 and the relative geometrical relationship is utilized to weaken the environmental interference.The corresponding detection result is utilized to initialize the mark tracking algorithm.The fDSST(fast Discriminative Scale Space Tracking)algorithm is adopted.For the down-view camera,the visual guide mark is detected based on the color and shape feature,and the spatial position of the landing target point is derived.An aerial video dataset is constructed and labeled for the training of the TinyYOLOv3 model.The detecting and tracking methods is verified under outdoor interference conditions.Thirdly,an autonomous landing strategy is designed to generate the UAV flight trajectory.Considering the difference between the detection range of the front-view camera and the down-view camera,the landing is divided into two stages.First,the front-view camera is used to guide the quadrotor to move close to the moving platform.Then,when the quadrotor is close to the moving platform,the down-view camera is used to guide the landing procedure of the UAV.According to the relative position information provided by the visual guidance,the flight path of the quadrotor is predicted in real time based on the constraints of the camera field of view.By ensuring the accuracy of the quadrotor trajectory tracking by the flight control system,the autonomous landing of the quadrotor on the moving platform is realized.Finally,the target detecting and tracking algorithm is implemented on the on-board embedded platform Jetson TX1,and the effectiveness of the target detecting and tracking strategy is verified by extensive experimental results.At the same time,the autonomous landing strategy is verified.The experimental results show that the quadrotor can achieve autonomous landing on the moving platform.