Research On The Algorithm For Detection And Recognition Of Foreign Objects Debris In Airport Runway

As a key part of the take-off and landing process of the aircraft,airport runway relates to the safety of all passengers.FOD(Foreign Object Debris)generally refers to some foreign material that may damage to the aircraft.The exsiting FOD detection synstems are difficult to popularize at domestic airports due to the technology monopoly and high purchase.At present,there is not a mature FOD detection system with independent property rights in our country.In this paper,we propose a video-based runway detection system on the airport shuttle car.This paper focuses on the algorithm of the FOD detection and recognition based on optical video images.In the FOD detection algorithm,this paper introduces human visual attention mechanism into the video-based runway FOD detection.Based on the traditional Itti's model,we further extracte brightness,color,edge and texture image feature,and then obtain feature significantly map by center-surround difference and normalization.We further synthesize the final visual saliency map according to the significant energy of each map,and finally using automatic threshold segmentation to realize the FOD detection.In the FOD recognition algorithm,in order to reduce the false alarm rate,it need to classificate foreign FOD.In this paper,we classificate and recognize FOD by Bagging classifier integration method.Firstly,we extract the color and texture feature as the recognition feature.And then we selected the support vector machine,BP neural network and k-Nearest Neighbor method as the based classifier to recognize FOD.We take the image of the FOD on the real runway.The experimental results show that our algorithm can distinguish the target and the background gray level in the synthesized saliency map,and is also well for the detection of small FOD.We select three typical FOD which from high risk,medium risk and low risk to test our recognition algorithm.The Cross-testing result shows that our algorithm based on classifier ensemble is better than anyone of the three classifiers...