无人机视觉辅助着舰算法研究

发布时间：2018-03-01 18:01

本文关键词： 无人机自主着陆视觉导航卡尔曼滤波　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：无人机体积小、成本低、能适应各种工作环境的特点使其广泛的应用于各个领域中,随着军用、民用以及商用的需求越来越大,关于无人机的研究有了显著的发展。在无人机着陆导航过程中,传统的惯导系统存在精度低、易受电子干扰等缺陷,基于机器视觉的导航方法越来越受到重视。本文着重研究了基于机器视觉的自主着陆算法,并通过无人机实地试验对算法的可用性作了验证,通过三维视景仿真对算法的准确性作了验证。并且将无损卡尔曼滤波应用于算法中,提高系统的精度和稳定性。首先,文章描述了无人机视觉导航系统的设计,着重研究了“姿态及位置确定”阶段,这一阶段分为图像处理模块和位姿估计模块,图像处理模块中,主要对图像进行了预处理、边缘检测、轮廓匹配等操作,确定找到合作地标后再使用光流法进行目标跟踪,以此方法来提高算法的效率和实时性。其次,文章研究了“姿态及位置确定”阶段中的第二个模块,即无人机视觉导航控制中的位姿估计模型。根据透视投影模型和坐标系相互之间的转换关系,建立一个数学模型,再加上实验获得有效特征点信息,就可以依据这个数学模型,可以推导出一个六分量的超定方程组,并使用奇异值分解的方法求解此方程组,并使用最小二乘法的方法估计出位姿信息的最优解。然后,文章使用大疆六旋翼无人机实践了自主着陆实地试验,操控无人机模拟飞行降落过程,将获得的视频资料用于设计的程序中,结果输出了一系列的位姿数据,即验证了算法的可执行性。并介绍了合成视景三维建模的一般流程,设计了一个经典算法,在理想的三维建模环境下,将无人机自身的位姿信息和实验所得位姿信息进行比对,误差在允许范围内,成功得验证了论文算法的正确性。最后,文章研究了无损卡尔曼滤波算法在无人机定位中的应用,包括异源传感器数据融合、和数据预估计,并且将算法设计采用MATLAB软件进行了仿真,结果证明卡尔曼滤波可以使位姿估计更加精确,系统更加稳定。
[Abstract]:Unmanned aerial vehicles (UAVs) are widely used in various fields because of their small size and low cost. With the increasing demand for military, civilian and commercial applications, UAVs can adapt to the characteristics of various working environments. The research on UAV has made remarkable progress. In the course of UAV landing navigation, the traditional inertial navigation system has some defects, such as low precision, easy to be interfered by electronic, etc. More and more attention has been paid to the navigation method based on machine vision. This paper focuses on the autonomous landing algorithm based on machine vision, and verifies the usability of the algorithm through the field test of UAV. The accuracy of the algorithm is verified by 3D visual simulation, and the lossless Kalman filter is applied to the algorithm to improve the accuracy and stability of the system. Firstly, the design of the vision navigation system for UAV is described. This stage is divided into image processing module and pose estimation module. In the image processing module, image preprocessing, edge detection, contour matching and other operations are mainly carried out in the image processing module. In order to improve the efficiency and real-time performance of the algorithm, optical flow method is used to track targets after finding cooperative landmarks. Secondly, the second module in the phase of "attitude and position determination" is studied in this paper. According to the transformation relationship between perspective projection model and coordinate system, a mathematical model is established, which can be based on the effective feature point information obtained by experiments, according to the transformation relationship between perspective projection model and coordinate system. A six-component system of overdetermined equations can be derived, and the singular value decomposition method is used to solve the equations, and the least square method is used to estimate the optimal solution of the position and attitude information. In this paper, an autonomous landing field experiment is carried out with DJ6 rotor-wing UAV, and the UAV is operated to simulate the flight landing process. The obtained video data are used in the design program, and a series of position and pose data are output. The algorithm is proved to be executable, and the general flow of 3D modeling of synthetic scene is introduced. A classical algorithm is designed to compare the position and pose information of UAV with that obtained from experiments in an ideal 3D modeling environment. The error is within the allowable range, and the correctness of the algorithm is verified successfully. Finally, the application of lossless Kalman filter algorithm in UAV positioning is studied, including data fusion of heterogenous sensors and data pre-estimation. The algorithm is simulated with MATLAB software. The results show that Kalman filter can make the position and pose estimation more accurate and the system more stable.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：V279

【参考文献】