| Torpedo has always been the core weapon in naval equipment,which has the advantages of high concealment,long-range and great power.The launching speed of a conventional torpedo is about 30m/s,which may lead to device failure,bouncing and sinking in the process of entering the water.The information provided by the target detection and tracking technology,such as attitude,position and velocity of the projectile,is the basic data to support the study of torpedo stability in the water.Therefore,it is important to design high-precision projectile detection and tracking algorithms.Conventional projectile position data is obtained by manual marking with camera software or sensor detection,which is poor in real-time and costly.With the development of computer vision,the target detection and tracking technology provide a convenient way to extract the localization of the projectile target.The conventional target detection and tracking algorithm cannot adapt to special phenomena such as water ripples and sudden changes of illumination during the process of projectile entry into the water.To address this problem,the thesis investigates how to suppress the effects of water ripples and sudden changes in illumination on the detection results of the cartridge,and how to improve the real-time detection on the basis of ensuring accuracy and stability.A projectile entry tracking software is designed to facilitate experimenters to obtain data such as projectile position and velocity.The main research contents and results are as follows:A self-calibrating Gaussian mixture model target detection algorithm based on projectile edge detection is designed(Canny CGMM).To address the interference problem of sudden illumination changes in the video on the detection of the projectile,a Gaussian mixture model(GMM)with a multisample background matching mechanism is used to reduce the sensitivity of illumination change detection.To address the problem of interference of water pattern disturbance on the detection of the projectile.A pre-frame calibration algorithm is designed to filter the water pattern foreground information in the non-elastic region.The bullet edge extraction and water surface detection strategies are proposed to filter out the water pattern foreground information in the projectile region.The experimental results show that the Canny CGMM algorithm outperforms the GMM algorithm in terms of resistance to illumination and water ripple interference,and the average pixel error is reduced by 12.1%.A kernel correlation filtered target tracking algorithm based on Kalman estimation is designed(Kalman KCF).To address the problem of the poor real-time performance of the projectile detection algorithm,the processing time is shortened by the frequency domain calculation strategy of the kernel correlation filter tracking algorithm(KCF).A Kalman model is designed to predict the position of the projectile for the problem of missing projectile tracking,and the prediction results are used to check whether the tracking is successful and whether the error is too large.The results of projectile detection by the Canny CGMM algorithm are used as the tracking template after calibration failure to improve the overall stability of the tracking algorithm.The experimental results show that the Kalman KCF algorithm reduces the average pixel error by 41.2% and improves the average FPS by 71.6% compared to the Canny CGMM algorithm.The traditional projectile position acquisition is manually marked by the camera’s software or acquired by the sensor in real-time,which is time-consuming and costly.The projectile position tracking software is designed by QT platform,equipped with Open CV vision library,integrated Canny CGMM target detection algorithm and Kalman KCF target tracking algorithm.The experimental results show that the projectile position,velocity and acceleration data obtained by the software are better than the manually marked data,which improves the data acquisition efficiency and reduces the experimental cost. |
PDF Full Text Request