Study Of Aircraft Landed By Vision Guidance And Airport Runway Detection Based On Deep Learning

With the development of deep learning and computer technology,aircraft landing guided by computer vision had became one of the research directions of intelligent civil aviation precision flight technology.Based on the target detection algorithm and visual positioning algorithm in the thesis,by collecting the ground image through the visual sensor,and identifies the airport runway area,determines the aircraft position synchronously,and realizes the computer vision to guide the aircraft landing.Through the research and application of YOLOv3 algorithm and ORB-SLAM algorithm,this thesis completes the airport runway recognition and aircraft positioning relative to the airport runway.At the same time,experiments verify the feasibility of computer vision guiding aircraft landing.Firstly,the basic theories of target detection algorithm based on deep learning and slam visual localization algorithm had studied and understood.Secondly,the airport runway detection model training and test experiment based on YOLOv3 were completed.By making600 VOC data sets including airport runway model pictures and real airport runway pictures,they were sent to YOLOv3 network to complete the airport runway detection model training.Using the airport runway detection model to predict the airport runway VOC test set,the average accuracy of the prediction results was 97.5%.Thirdly,the research and improvement of ORB-SLAM algorithm had completed.In the Ba solution part of the back-end of ORB-SLAM system,Schur elimination was added to accelerate the calculation,and Ceres optimization library was used to replace g2 o optimization library to complete the improvement of the back-end of ORB-SLAM system.Using ORB-SLAM and improved ORB-SLAM to solve the official data set respectively and complete the experimental data analysis,the results show that the trajectory error of the improved ORB-SLAM output was lower and the error fluctuation was more stable.Finally,through the preparation of the previous theoretical and experimental basic data,it had completed the positioning experiment of the aircraft relative to the airport runway in the thesis,and puts forward the voice guidance scheme on this basis.The depth camera was used to approach the entrance of the airport runway model at a constant speed to simulate the aircraft landing scene.The color map and depth map of the airport runway model were collected by the camera to complete the production of tum data set.The improved ORB-SLAM algorithm was used to solve the tum data set,output the camera trajectory,fit the trajectory,compare it with the 3 ° standard glide line,and complete the error analysis.The results show that the trajectory error was basically consistent with the error of the official data set.By fitting the trajectory and comparing the standard glide line,according to the coordinate deviation of eight typical trajectories,eight voice command suggestions were put forward to complete the voice guided landing.The final results show that computer vision guided aircraft landing was feasible.