Vision-based Design For Autonomous UAV Landing

Rotor UAVs are widely used in military reconnaissance,agricultural plant protection,and power inspection due to their small size,simple structure,and vertical take-off and landing.With the development of artificial intelligence and microprocessor hardware systems,the intelligent flight of UAVs has become a current hot topic,and autonomous landing of UAVs,as an important part of achieving intelligent flight,has also received widespread attention.Among many navigation systems,visual navigation has the advantages of strong anti-interference ability and low maintenance cost,which is very suitable for the precise landing of UAVs.Therefore,this paper conducts an in-depth study on the vision-based autonomous landing algorithm for UAVs.In order to enable the UAV to enter the landing stage from a higher altitude and have a higher accuracy at the end of the landing,this paper proposes a three-stage UAV landing plan that includes high altitude,low altitude and predicted geometric feature points.The three algorithms of high altitude,low altitude and predicted geometric feature points can be automatically switched according to the altitude information obtained by the barometer.In the high-altitude visual sign recognition and landing stage,in order to enable the UAV to detect visual signs from a higher altitude and meet the real-time requirements of the algorithm,this paper proposes a lightweight YOLO-UAV target detection algorithm to achieve large-scale sign search and Recognition.First,the visual image obtained by the airborne camera is input to the lightweight YOLO-UAV target detection network for feature extraction;then,the candidate prediction frame is output through the algorithm;finally,the prediction frame with the highest confidence score is output as the final prediction frame.And determine the position of the visual sign according to the information of the output prediction frame.In the low-altitude visual sign recognition and pose estimation stage,in order to better recognize the visual sign,this paper proposes the FAST-UAV feature point detection algorithm.Firstly,multiple FAST-UAV candidate feature points are detected by the algorithm;secondly,the angle of FAST-UAV candidate feature points is limited by the angle particularity of the visual landmark feature points,and 12 feature points of the visual landmark are obtained;,After the identification mark contour extraction and the feature point label,the center point coordinates and main direction of the identification mark are calculated,and the feature point classification is completed to complete the feature matching;finally,the homography matrix and the "H" shaped mark calculated according to the feature matching The geometric characteristics of the UAV can be calculated to calculate the pose parameters of the UAV relative to the visual sign,and prepare for the autonomous landing of the UAV.In the final blind landing stage before landing,in order to solve the problem of the UAV being forced to blindly land due to the loss of feature points,a predictive geometric feature point algorithm is proposed.This algorithm uses the idea of geometrically predicting feature points to supplement the missing feature points.To achieve the final stage of precise landing and the entire autonomous landing control of the UAV.Finally,the algorithm was verified experimentally using the rotary-wing UAV experimental platform and simulation experimental platform,and the effectiveness,accuracy and time performance of the three-stage landing algorithm were tested and analyzed.The experimental results showed that the algorithm can meet the requirements of autonomous landing of UAVs Claim.