| In recent years,the technology of quadrotor UAV has become increasingly mature.Its small size,flexible operation and stable flight have been widely used in various fields.Among them,the functions of autonomous endurance,logistics distribution and safe recovery of UAV need the support of UAV landing technology.Therefore,autonomous landing is one of the key technologies of UAV.The traditional UAV landing technology based on GPS and inertial navigation is not suitable for micro UAV due to the constraints of flight environment and hardware conditions.Visual guidance methods provide a new solution for micro UAV landing,which can meet the needs of low-cost,high reliability,and flexible flight.Based on this,this paper proposes a micro quadrotor UAV autonomous landing method based on visual guidance,including landmark design and detection,UAV position estimation,landing guidance strategy,etc.The main contents are as follows:(1)A double-layer nested Aruco landmark and its detection algorithm are proposed to address the problem of difficult recognition of landmarks in complex environments due to occlusion,and changes in the size of landmark images captured by airborne cameras as flight altitude decreases.Firstly,based on the Aruco,a double-layer nested landmark is designed.The outer Aruco is used to guide the UAV at high altitude towards the landmark,while the inner Aruco is used to calibrate the position of the UAV at close range;Secondly,the landmark dataset containing different scene is made,and the YOLOv4 deep learning algorithm is used to replace the traditional Aruco detection algorithm to adapt to the changes of the environment such as the occlusion of the landmark.Finally,in order to improve the detection accuracy of small target landmarks during high-altitude flight of UAV,normalized Wasserstein distance is used to calculate the similarity between adjacent small targets,replacing the traditional intersection and union ratio measurement method as a measure of redundant boundary box screening in non maximum suppression.The experimental results show that the improved landmark detection algorithm can effectively detect the designed landmark and reduce error detection of high-altitude small target landmark images.The network model is trained using the produced landmark dataset,with a detection accuracy mAP of 98.22%and a detection speed of 72 FPS,meeting the real-time requirements of UAV visual landmark detection.(2)Aiming at the problem that the height estimation accuracy of micro UAV based on monocular vision is not high,a method of combining inertial measurement unit and visual height is proposed to improve the accuracy of UAV flight height estimation.Firstly,based on the collected landmark images,combined with perspective projection algorithm and triangle similarity principle,the UAV is positioned to obtain the relative position relationship between the landmark and the UAV,Then,in order to improve the estimation accuracy of flight height value,the height estimated based on image is subjected to secondary differentiation and fused with accelerometer data.The obtained acceleration value is subjected to secondary integration,Kalman filtering,and then complementary fusion with the image height value;Finally,based on the feature point information obtained from the landmark detection algorithm and the corrected UAV flight altitude data,combined with visual servo algorithm,the position between the UAV and the landmark is adjusted to continuously reduce their position deviation,in order to control the UAV to approach the landmark.The experimental results show that the proposed height numerical correction method can effectively improve the problem of image height data lag,especially in the near ground landing stage.In addition,flight experimental results show that using the UAV visual position estimation algorithm,the position deviation between the final landing point of the UAV and the landmark center point in the X direction is about 8cm,and the position deviation in the Y direction is about 5cm,which meets the requirements of the UAV landing task.(3)A simple and low-cost landmark image recapture strategy is proposed to address the sudden loss of landmark images when UAV is landing on a static target.When the landmark image is lost in the field of view of the airborne camera,firstly,the UAV hovers in place and attempts to capture the landmark image again;If the landmark image is still not detected,it will rise to the initial altitude,increase the field of view of the airborne camera,capture the landmark image again,and continue the UAV landing task;When the UAV is landing at a safe distance of 20cm from the ground,the UAV is locked,the rotors stop rotating,and ultimately relies on gravity to land on the landmark.The simulation results show that the proposed landmark image recapture strategy can solve the problem of landmark image loss to a certain extent. |
