Research On Recognition And Tracking Method Of Landing Marker Of UAV

With the widespread use of drones,the autonomous landing ability of drones has become increasingly important,and accurate and safe autonomous landing directly affects the success or failure of their missions.Vision-assisted autonomous landing is an economical,efficient,and robust solution with strong anti-interference capabilities.Therefore,this paper conducts research on the identification and tracking methods of drone landing markers based on the application scenario of vision-assisted drone autonomous landing.The specific work is as follows:(1)A landing marker was designed,and a dataset of landing markers was created.First,a landing marker was designed,and then the dataset of landing marker s was created through operations such as image collection,image enhancement,and image annotation.(2)An improved YOLOv5 algorithm was proposed for drone landing marker identification.To address the problem of small targets encountered when identifying ground markers with drones and the high computational cost of the algorithm,the YOLOv5 algorithm was improved.Firstly,the feature extraction ability of the backbone network was improved by adding a tiny object detection layer and introducing an efficient channel attention mechanism,thereby improving the accuracy of small target recognition.Secondly,the ordinary convolution in the CBL module was replaced to improve the running speed of the algorithm.Then,the SIOU loss function was used as the bounding box loss function to improve the localization accuracy.Finally,the improved YOLOv5 algorithm was verified,and the results showed that the improved YOLOv5 algorithm could effectively detect smaller landing markers while improving detection speed.(3)An improved KCF algorithm was proposed for drone landing marker tracking.To address the problems of scale changes and occlusion encountered when tracking landing markers with drones,the KCF algorithm was improved.Firstly,an adaptive multi-scale strategy was designed to deal with the scale changes of tracking markers in the field of view.Secondly,a re-detection mechanism based on the improved YOLOv5 algorithm was proposed as the confidence indicator to improve the tracking success rate of landing markers under occlusion.Finally,the improved KCF algorithm was verified,and the results showed that the improved KCF algorithm could effectively improve the tracking effect of landing markers under scale changes and occlusion.(4)The landing marker identification and tracking model was deployed on an embedded platform,and a visualization software for landing marker identification and tracking was designed and developed.First,the landing marker identification and tracking algorithm were deployed on Jetson Nano B01,and model acceleration was performed to improve the model’s running speed.Secondly,the landing marker identification and tracking visualization software was built using Qt software,with functions such as image pre-processing,real-time identification and tracking of landing markers,display,and saving.Finally,the built visualization software was tested,verifying the effectiveness and applicability of the software.In summary,this paper proposes identification and tracking algorithms for drone landing markers,and deploys the landing marker identification and tracking algorithm on an embedded platform,as well as developing visualization software for landing marker identification and tracking.