Research And Implementation Of Airport Runway Foreign Object Debris Detection System

In recent years,with the rapid development of the air transport industry and the increasing of people's consumption level,more and more people choose to travel by plane.The blowout of air passenger traffic has brought opportunities to the air transport industry as well as great challenges.In these challenges,the safety of travel is particularly noteworthy.FOD(Foreign Object Debris)refers to foreign objects on the airport runway that may cause damage to the aircraft.FOD will bring huge security risks to the aircraft's takeoff and landing.In response to this problem,this thesis focuses on the related technologies of the foreign object detection system of the airport runway.Based on the runway video images collected by the highdefinition camera,we have initially implemented a system that can detect the invasion of foreign objects in real time and classify the foreign objects to initiate different levels of security alerts.First,we research image processing algorithms including image graying,image noise reduction,and image enhancement.The advantages and disadvantages of commonly used image noise reduction algorithms are analyzed and verified through simulation experiments,which provide support for the subsequent image preprocessing steps of the foreign object detection algorithm.Next,we research commonly used target detection algorithms and summarize the advantages and disadvantages of each algorithm.Considering the system's need for stability and efficiency,we choose the frame difference method as the system's target detection algorithm.Because the calculation process of the frame difference method is simple and efficient.But the traditional frame difference method has the defect of losing the target edge information.We improve the traditional frame difference method by adaptive binarization threshold and target contour correction based on Canny edge detection algorithm.It is verified through simulation experiments that the improved target detection algorithm can extract target information more completely,and can effectively detect the invasion of foreign objects.Then,this thesis introduces the basic structure of the convolutional neural network model and transfer learning.We construct a foreign object classification model by fine-tuning the Res Net which is a convolutional neural network model with a top-5 error rate as low as3.5% on the Image Net data set.We trained the classification model by using the image data sets which were obtained by the crawler technology.The accuracy of the classification model in the verification set has reached more than 95%.Finally,according to the functional requirements of the FOD detection system mentioned in the advisory notice issued by the Federal Aviation Administration,we design and implement a software system that includes a PC-side master control module,a mobile-side alarm processing module,and an image processing module.We designed test cases and tested the system in detail.The results of test show that the system is stable and reliable.