Research And Realization Of Gun Barrel Orientation Measurement Method Based On YOLOv5 And Computer Vision

The fire command and control system are the core of the whole artillery system.Its static and dynamic gun barrel pointing measurement accuracy is the focus of the performance index of the fire control system,which is also an important reference for artillery shooting test,equipment status characterization,system fault prediction and diagnosis.Therefore,aiming at the problems existing in the current gun barrel measurement method,such as complex measurement method,poor target adaptability,real-time dynamic measurement and low accuracy.This thesis studies the gun barrel pointing measurement system based on YOLOv5 and computer vision measurement method.What’s more,the correctness and practicality of the method are verified by experiments.Firstly,the system model building and network structure design are studied.Based on the position and orientation relationship between the gun and the barrel target in the field of view,the gun barrel positioning model is designed.The target feature points are obtained by combining straight lines and curved lines,and converted into P3 P problem to establish the geometric positioning model.Besides,the optimal solution of the barrel target model in different field of view angles is calculated.In addition,the camera imaging model under radial and tangential distortion is analyzed.By comparing different camera calibration methods,the camera calibration technology required by the task is analyzed and the initial value of the model is calculated according to the camera’s position and attitude of the field of view.Besides,the optimal Kalman gain is derived and the feasibility of the camera calibration method based on chessboard in the actual measurement task is verified by experiments.What’s more,the target recognition and tracking model based on YOLOv5 is designed,the principle and process of gun barrel target tracking and recognition are described,the target feature extraction method is studied,and the loss function of the training model is improved on the basis of the original network framework.Secondly,the static and dynamic position and attitude measurement methods of gun barrel under linear target are studied.The linear barrel target model is trained.Besides,the labels file is obtained using the MAKE SENSE data production environment,regression strategy of the barrel target frame is studied.What’s more,the convergence speed of the model is improved,and the model training parameters of the linear barrel target are given.The principle of obtaining the reference value of the linear gun barrel model and the verification method of the measurement results are studied.The static and dynamic measurement experiments of the gun barrel under the linear target condition are carried out.In the data processing and analysis,the judgment basis of outlier elimination is determined,and the experimental data is optimized using the data fitting algorithm based on the support vector machine.Finally,the measurement results of the gun barrel pointing under different motion modes are visually characterized.The correctness and effectiveness of this method are verified in the case of linear target.Thirdly,in the case of linear target,considering that the accuracy of gun barrel pointing measurement is affected by the target nonlinear distribution,a nonlinear barrel target identification and pose solution method based on YOLOv5-Kalman is proposed.Firstly,based on the linear barrel target training model,the model training links such as model preprocessing,data enhancement,training optimization and post-processing are improved,and the cosine annealing algorithm is used to adjust the model parameters to improve the model validation accuracy,and the training effect of the model is characterized by visual processing.Then,the improvement effect of different algorithms on YOLOv5 barrel target model in the case of nonlinear target is analyzed,and the Kalman filter gain with the smallest model error is derived.Then the static and dynamic measurement experiments of the designed YOLOv5-Kalman model are carried out,and the target tracking and recognition accuracy of the model and the position and attitude calculation error of the gun barrel are analyzed,which verifies the correctness and adaptability of the model to improve the gun barrel pointing measurement under the condition of nonlinear target.Finally,according to the research content of this thesis,a set of gun barrel pointing measurement system is developed.Firstly,the static and dynamic measurement modules are designed,and the static and dynamic measurement processes under different target conditions are given.The specific principles and processes of the camera target stator module,the camera field of view coordinate positioning submodule,and the multi-machine synchronous trigger sub-module required in the visual measurement module are studied.Finally,the software interface of the gun barrel pointing measurement system after design is given,including the main interface of the main control computer module,the functional links and distribution status interface of the communication networking module,and the functional distribution,data distribution and data curve interface of the barrel pointing measurement module.