Research On Monitoring Technology For FOD Of Airport Runway Based On Machine Vision

In recent years, with the rapid development of aeronautial science and technology, the role played by civil aviation transportation is more and more important in human life, combat effectiveness of military aviation also directly determines the course of national defense construction, therefore, to guarantee that civil and military aircrafts flight safely has great practical significance. However, Foreign Object Debris(FOD) on the airport runway poses serious threaten to the security of aircraft landing, and has resulted in heavy losses. In allusion to this problem, on the basis of the theories of machine vision and digital image processing,the key technologies of FOD monitoring system are studied focusly in this thesis, automatic detection and recognition for invaded FOD are realized initially, the software prototype of FOD monitoring system based on machine vision is designed and achieved.A deep research analysised the principles and advantages and disadvantages of common moving target detection algorithms, for the particularity of the runway environment is relatively single, the average background subtraction algorithm based on adaptive background updating for FOD detection is used. A relatively reliable initial background is established through extracting the average gray value of first N-frame images, the minus operation is used between current frame and the background image to receive a foreground iamge and then get it’s binarized image, the interference from non-FOD targets like aircrafts is eliminated by contour area abandoning, finally, carring out morphological operation on the binarized image and a more accurate test result is obtained. In order to adapt to the changing environment in real time, the background image is weighted updating processed duly; taking into account the influence of big illumination changes from the spectial weather such as dark clouds, a light detecting operation is conducted on the current frame and an adaptive update rate is used to update the background image, the simulation experiments verify the effectiveness of the algorithm.Because of the types of FOD are very various, several typical FODs are selected as objects of study and to be analyzed their characteristics, the rotation invariant pattern LBP operator is used to extract texture feature values of FOD as a basis for classification and recognition, select the statistical histogram of FOD’s texture feature as input of classifier;multi-classifier Support Vector Machine(SVM) based on two-classification is used for classifing, offline training of SVM is completed through one-to-one classification method,and online targets classification is executed by "referendum rules". A large number of positive and negative samples is selected for simulation, experimental results show that correct rate for target recognition can meet the requirements of this system.Software prototype of FOD monitoring system is designed and completed, whichincludes six functional modules such as user login, video capturing, video storaging, runway area setting, FOD detecting and recognizing, and alarm managing. Once a FOD invades the monitoring window, detection and recognition can be carried out for it, at the same time, an alarm information is been given according to the threat level of FOD.