Research On Airport Runway FOD Detection Based On Vehicle Camera

Airport runway Foreign Object Debris（FOD）generally refers to some foreign substances that may damage aircraft or systems,causing great harm to aircraft.As a deadly killer in the take-off and landing stage,it can cause flight delay,takeoff interruption,and even endanger passengers’ life safety.At present,artificial detection is mostly used to detect foreign body in airport runway,which has low efficiency and poor reliability.In order to change this situation,both at home and abroad have begun FOD automatic detection research,and some of the results have been tested at some airports.mainly based on millimeter wave radar tower supplemented,machine vision detection mechanisms,however,high cost of millimeter wave radar,and a certain height of tower system by Atc height limitation and space limitation,often hard to promote and popularize,Therefore,it is urgent to develop a flexible,efficient,low-cost and easy to popularize FOD detection system for airport runway.In this paper,vehicle camera is used to collect data and image processing technology is used to detect foreign body in airport runway.the main structure extraction and foreign body detection of airport runway are studied.First of all,considering that the asphalt runway will interfere with the detection of small foreign bodies,it is necessary to extract the main structure of the airport runway.Therefore,the weighted least squares method（WLS）and Relative Total Variation Model（RTV）are analyzed and studied respectively in this paper.And the problems of the two main structure extraction algorithms are improved.Then,the advantages and disadvantages of the two algorithms are compared and analyzed in terms of main structure extraction efficiency and extraction speed,and the improved relative total variation model is selected to extract the main structure of the airport runway.Finally,based on the main structure extraction,the improved Pixel-based Adaptive segmentation（PBAS）algorithm is used to detect foreign bodies on the runway.To sum up,the contributions of this paper are as follows:（1）An improved weighted least squares image main structure extraction algorithm is proposed.Firstly,the weighted least square method cannot remove the oscillating details（such as texture）with low semantic contribution,small scale and high contrast due to the influence of gradient size,a new smoothness weight for suppressing image texture is proposed in this paper,and the validity of this weight is verified.Then,in order to solve the problem of high computational cost of solving large sparse linear matrix equation in the process of optimizing global objective function,this paper adopts algebraic multiple grid algorithm as the pre-processing operator of conjugate gradient method to accelerate the extraction speed of main structure.（2）An improved main structure extraction algorithm based on relative total variation model is proposed.In the relative total variation model,texture and main structure are distinguished mainly by window inherent variation,but the edge of main structure extracted by this algorithm is easy to be blurred when texture is filtered.In this paper,by improving the weight function of the model,the gradient relation between the surrounding pixels and the center pixels is introduced into the weight function,so as to improve the contrast of the main structure and texture.The improved algorithm does not blur the edges of the main structure when filtering the image texture,and the difference between the main structure and the texture is more significant.（3）On the basis of the main structure extraction of the runway,foreign body detection on the runway is carried out,and a foreign body detection algorithm based on background difference is proposed.The weight mechanism is introduced into the background model samples to improve the expression ability of the background model.In the foreground classification stage,counting mechanism is introduced to solve the problem of incomplete detection of stationary or low-speed moving targets.In the weight updating stage,the reward and punishment strategies are adopted to increase the weight value of active samples and decrease the weight value of inactive samples in the background model samples.It improves the ability of track foreign body detection under complex background.The experimental results show that the main structure extraction of runway can reduce the interference of runway texture to foreign body detection.Then the improved background difference algorithm can accurately and effectively detect foreign bodies in the runway.