Aircraft Landing Gear Status Monitoring Based On Deep Learning

Real-time aircraft landing gear status monitoring aims to detect the aircraft target and the landing gear and to obtain angle information,which is a necessary condition for the safe landing of aircraft.Condition monitoring of the aircraft landing gear is a prerequisite for the safe landing of the aircraft.Traditional manual observation has a strong subjectivity.In recent years,target detection models based on deep learning and pose estimation methods based on a single RGB image have made great progress.The existing landing gear status monitoring is reported to the ground tower by the aircraft itself.In this paper,a method to monitor the landing gear status directly from the ground is proposed based on monocular camera image and three-dimensional model of the aircraft.Inputting a single RGB image of an aircraft,the key points of landing gears are obtained through the target detection module.The key points of the fuselage are voted by the vector field network after extraction and scale normalization of the pixels inside the aircraft prediction box.Knowing the pixel position of the key points and the constraints on the aircraft,the angle between landing gear and fuselage plane can be calculated even without depth information.The vector field loss function is improved based on the distance between pixels and key points,and synthetic datasets of aircraft with different angle landing gears are created to verify the validity of the proposed algorithm.The experimental results show that the algorithm’s mean error for the landing gears is less than 8 degrees on the proposed datasets.The main innovative research work and research results of this paper are as follows:（1）The monitoring objectives,application characteristics and performance evaluation criteria of landing gear condition monitoring are analyzed.（2）In order to solve the problem that the scale of photographed aircraft targets varies greatly,this paper designs a target extraction and normalization module based on YOLO-V4.The function is to extract the image of the aircraft target area,normalize the image size,and reduce the pixel position error of the fuselage key points caused by the scale change.（3）To solve the problem that the landing gear can move,the aircraft is divided into two parts of the fuselage and landing gear to detect the key points.The landing gear features prominently and is detected by the target extraction module.The rest of the aircraft can be treated as rigid bodies and detected using a robust pixel-level voting network called PVNet.（4）To solve the problem that there is no landing gear Angle data set,we use the aircraft CAD model to make the aircraft simulation data set with different landing gear angles,which is suitable for the algorithm in this paper.（5）To solve the difficulty of obtaining depth information of aircraft,we designed CAL module,which can directly calculate landing gear Angle from the key pixel position of two-dimensional image of known aircraft model structure.The CAL module avoids the regression of aircraft space attitude and improves the measurement accuracy.（6）The performance of the proposed landing gear condition monitoring algorithm was evaluated by ablation and comparison experiments of each module and innovation of the algorithm.