Research On Self Disturbance Rejection Landing Method Of Visually Assisted UAV In Complex Environment

The increasing demand for drones to exhibit greater intelligence is propelled by the continuous expansion and deepening of their applications across various industries.Autonomous landing of UAVs is considered a prerequisite for realizing the intelligence of UAVs.The vigorous development of computer vision offers an efficient and low-cost solution for UAVs to acquire landing information.However,poor visual acquisition conditions can lead to the failure of detecting landing markers,and stable control of UAVs is greatly restricted by unpredictable external timevarying wind disturbance,hindering the rapid and accurate detection of UAVs during landing due to the complex landing environment of UAVs.In order to solve the above problems,this thesis proposes a vision-assisted antidisturbance landing method for quadrotor UAVs in complex environments,separates and processes different interference sources,and conducts targeted research in the visual positioning part and flight control part respectively.The main contributions as follows:(1)An improved Lucy-Richardson deblurring algorithm combined with motion information is proposed to address the motion blur problem of the marker image.The real-time motion information of the UAV is utilized to derive the blur kernel,and a Gaussian filter is introduced to achieve the separation processing of high and low frequency information.To address the poor detection ability of Yolov5 s for small targets,a SLME-Yolov5s(Small Landing Marker Enhanced Yolov5s)detection algorithm with enhanced small landing mark detection capability is proposed.The high-level feature map is directional clipped in the Backbone part to reduce computational complexity and increase the number of small targets detected by a small target detection head to obtain more positioning information.In addition,Bidirectional Feature Pyramid Network(Bi FPN)and Spatial Pyramid Pooling Fast(SPPF)are introduced in the Neck part to achieve better feature fusion.The detection experiment results on the landing marker dataset show that SLME-Yolov5 s has stronger small target detection ability and faster inference speed.(2)Aiming at the problem of unpredictable time-varying wind disturbance during the landing process,an improved Fractional Order Linear Active Disturbance Rejection Controller(FOLADRC)with a fractional calculus is proposed.The fractional calculus is introduced into the linear active disturbance rejection controller,and the Fractional Order Linear State Error Feedback(FOLSEF)is used to replace the Linear State Error Feedback(LSEF)in the LADRC.The fractional derivative is introduced into the differential term,which better adapts to the system’s dynamic changes and external disturbances.While retaining the powerful estimation capability of internal and external disturbances of LADRC,FOLADRC is better suited to the system’s dynamic changes and external disturbances.Simulations conducted in Simulink demonstrate that FOLADRC has faster response speed and stronger antiinterference ability.(3)The effectiveness of the proposed algorithm was further verified through ROS simulation tests and physical experiments.In the ROS simulation test,a virtual environment that highly restores the real scene was constructed using the Gazebo simulation platform,where a comprehensive functional test and performance evaluation of the system were conducted.Meanwhile,in the physical experiment,the algorithm was deployed on a quadrotor UAV,where the robustness and stability of the system were tested through actual landing control.