Design And Implementation Of Airport Scene Aircraft Detection And Tracking System Based On Deep Learning

With the rapid development of China's civil aviation industry,the airport mobile area is expanding day by day,and the traffic situation becomes more complex,which brings great pressure to tower control.In order to alleviate the control pressure and effectively guarantee the flight safety of the scene,the air traffic control part often uses the scene surveillance radar and multi-point positioning system to assist the tower controllers to grasp the scene traffic situation in time.However,this kind of surveillance system has high equipment costs and maintenance costs,which can not be borne by small and medium-sized airports.The airport scene aircraft detection and tracking system based on in-depth learning has the advantages of low cost small blind area and intuitive display.It has become a better alternative system,and can also be applied to remote command and anti-runway intrusion of civil aviation airports.This paper designs and implements an airport scene aircraft detection and tracking system,which can assist tower controllers to clearly and intuitively monitor the movement status of airport scene aircraft,improve the efficiency of surveillance work,and ensure the safe operation of the air traffic control department.Firstly,this paper applies YOLO V3(You Only Look Once v3)algorithm based on deep learning model to the field of airport surface aircraft detection.Secondly,Kalman filter and Hungarian algorithm are combined to realize the tracking of video aircraft.Finally,the aircraft tracking information is fused with ADS-B(Automatic Dependent Surveillance-Broadcast)data to display the tracking frame and corresponding flight information on the scene video plane.The main work of this paper is as follows:(1)The YOLO V3 algorithm with high accuracy and high speed is used to detect the aircraft on the scene,and the Aeroplane data set of the aircraft on the scene is constructed.In view of the shortcomings of YOLO V3 in small target detection and occlusion,the optimized YOLO V3 algorithm is optimized.The experimental results show that the optimized YOLO V3 algorithm has better detection performance in small target and aircraft occlusion detection.(2)Aircraft tracking.On the basis of the results of the first frame of video,Kalman filter is used to detect the position and motion state of the aircraft and predict the position of the aircraft in the next frame.Combining Hungarian algorithm and IOU(Intersection over Union)measurement method,the data association and real-time aircraft tracking of the two frames are realized.(3)Data fusion,which maps the pixel coordinates of the aircraft in the video to the longitude and latitude coordinates of the aircraft provided by the civil aviation surveillance system ADS-B,realizes the fusion of airport scene video to display the tracking frame and flight information of the aircraft.