Target Detection Method Based On Statistical Learning Is Applied To Fod Dection For Airport Runway

FOD detection for airport runway is an emerging issue and has been proposed in recent years. All researches about FOD detection include millimeter wave radar technology and image processing technology. If we adopt millimeter wave radar technology, the cost of FOD detection will be very high; the size requirements of FOD will be high; the requirements of hardware device will be very high. Moreover, China doesn’t have mature and relevant commercial systems. If we adopt image processing technology, the cost of FOD detection will be relatively low; the requirements of hardware device will be relatively low. Moreover, image monitoring system has been installed in the vicinity of runway in some airports. Thus, in this paper, we adopt image processing technology to achieve FOD detection.In this paper, the target detection method based on statistical learning is applied to FOD detection system. We make use of mature face detection system; keep Adaboost classifier method commonly used in face detection system; improve or exclude features commonly used in face detection system; find a suitable feature for FOD detection system. First we introduce LBP feature commonly used in face detection system; apply LBP feature to FOD detection system; verify the feasibility of LBP feature through experimental results. The experimental results show that LBP feature is not feasible. According to the characteristics of airport runway pictures, we propose a new histogram feature which is based on SUSAN feature. Then the experimental results show that new feature is feasible. Secondly in the pictures to de detected, there are a lot of false alarms around the runway lines. To solve the problem we propose a way based on image segmentation to remove the false alarms around the runway lines. We find the positions of the runway lines through marking edge points. According to the positions of the runway lines, we divide the pictures into parts containing the runway lines and parts without containing the runway lines, respectively make use of target detection based on Kirsch feature and statistical learning. Finally we introduce the principle and training processes of Adaboost classifier.The main contributions of this paper are summarized in the following three points:(1)A new histogram feature based on SUSAN feature is proposedWe propose a new histogram feature which is based on SUSAN feature and apply the new feature to FOD detection system. Through the experimental results show that the new feature is feasible.(2) A line detection method based on marking edge points is used to detect line.We remove the false alarms round the runway lines by segmenting images. Actually we segment image through the positions of the runway lines. So the primary task is to determine the positions of the runway lines. The paper presents a line detection method based on marking edge points.Compared with the line detection method based on Hough transform, the method is not only simple and feasible, but also determines the positions of end points.(3)Target detection method based on edge feature and statistical learning is combined.After we segment the pictures, we will combine the target detection method based on edge feature and statistical learning and respectively process different parts. The experimental results show that the method is feasible.