The Real-time Measurement Method Of Ship-borne Aircraft Position Based On Machine Vision

Real-time detection of ship-borne aircraft position is of great significance to the motion control,trajectory planning and collision prevention and safety of ship-borne aircraft on deck.The traditional ship-borne aircraft deck scheduling mainly relies on manual judgment of the position and course angle of the ship-borne aircraft,can not get accurate data in order to use automation equipment to achieve real-time motion control and trajectory planning,easy because of operator negligence and fatigue collision accident.The purpose of this paper is to simulate the monitoring system on the mother ship by using fixed position and angle camera,to observe the motion and stationary multiple ship-borne aircraft on the mothership deck by using multiple cameras,and to give real-time position by machine vision.The main steps of ship-borne aircraft position detection proposed in this paper can be divided into two parts: target recognition and segmentation,and position detection algorithm under single target.The main purpose of target detection and segmentation is to obtain multiple pictures from multiple cameras and divide them into sub-images according to the ship-borne aircraft target.In this thesis,the target identification and segmentation are studied and experimented,under the premise of extracting edge characteristics,the k-means clustering algorithm,DBSCAN clustering algorithm are mainly studied,and a axis pixel statistics method is proposed by edge information distribution characteristics.The R-CNN network has the best recognition.The single-target position detection algorithm is mainly responsible for giving the target’s position in real time according to the results of target detection and segmentation.Based on the template matching algorithm,this thesis innovatively puts forward an edge information and wireframe template matching algorithm.Starting from the algorithm,the algorithm not only has the advantages of template matching anti-noise interference and accurate results,but also ensures that its running speed meets the real-time requirements by background subtract,closed operation and contour extraction,bird’s eye view transformation de-re-estimation and initial solution estimation,and distance cost-based matching algorithm.Based on the above algorithm,this thesis writes the test software and constructs the physical verification environment to test the algorithm performance.In order to get better real-time performance,this paper uses CUDA to accelerate the algorithm under Nvidia 3080,and designs the corresponding parallel algorithm and pipeline pair to optimize.At the same time,a physical environment of 1:70 is constructed to simulate the actual mothership situation to test the algorithm,the algorithm recognition rate of more than 95%,accuracy of 2.5degrees,5mm(1.5% of the ship-borne airport),speed of more than 3.5Hz(4 cameras,8ship-borne aircraft targets).Experiments show that the algorithm can do a good job of real-time detection of ship-borne aircraft position.