【作者】 张伟；

【导师】 马珺；

【作者基本信息】 太原理工大学 ， 控制科学与工程， 2019， 硕士

【摘要】 随着科技的日益革新,无人机已广泛地应用于军事、民用等各个领域。无人机自主着陆是无人机飞行过程中的重要阶段。由于经典GPS/INS组合导航系统已经无法满足当前复杂的飞行环境要求,很难在各种情况下为无人机自主飞行提供精准的导航信息,故基于视觉的无人机着陆技术成为无人机控制领域的研究热点之一,亦弥补了国内无人机自主着陆于移动平台技术的空缺。本文以四旋翼无人机为实验平台,以机载摄像机所拍摄的序列图像为研究对象,综合运用计算机视觉、数字图像处理、透视投影等技术和方法,对无人机自主着陆阶段中的着陆标志跟踪、特征信息提取、无人机相对位姿参数的估计等问题展开了深入的研究,主要探讨了其基于视觉的目标跟踪与位姿解算问题。在目标跟踪问题上,先设计了“H”形的着陆目标图像,提取SIFT特征信息并匹配,采用RANSAC算法进一步完善匹配结果。结合Camshift跟踪算法和粒子滤波跟踪算法的优点,提出了一种融合目标颜色特征、SIFT特征和纹理特征的粒子滤波跟踪算法,详细描述了其工作流程。首先依据SIFT匹配结果确定跟踪窗口的初始位置和大小,然后以Camshift算法优化粒子的传播,自适应调节跟踪窗口的大小与方向,最后估计跟踪目标的状态。同时还提出了三种特征信息的融合策略,以及针对粒子退化现象引入了粒子重采样技术。实验验证了本文跟踪算法在目标被遮挡和相似颜色干扰的情况下均有效可行。在位姿解算问题上,对机载摄像机采集到的被跟踪区域的图像经过阈值分割、中值滤波、角点检测等图像处理流程后,得到“H”形着陆标志特征点在图像上的映射点,再依据透视投影模型和坐标系相互之间的转换关系建立无人机位姿估计模型,最后采用奇异值分解法解算位姿信息的最优解。实验中,将无人机自身的位姿信息和实验所得位姿信息进行比对,其误差在允许范围内,验证了本文算法的正确性。更多还原

【Abstract】 With the continuous innovation of science and technology,unmanned aerial vehicle(UAV)have been widely used in various fields such as military and civilian.Autonomous landing of UAV is an important stage in the process of UAV flight.Because the classic GPS/INS integrated navigation system is difficult to provide accurate navigation information for the autonomous flight of UAV in various situations,the vision-based UAV landing technology has become one of the research focuses in the field of UAV control,and it also makes up for the vacancy of domestic UAV autonomous landing on mobile platform technology.In this paper,the four-rotor UAV is used as the experimental platform,and the sequence images taken by the airborne camera are taken as research objects.The techniques and methods such as computer vision,digital image processing and perspective projection are used to land in the autonomous landing stage of UAV.The problem of marker tracking,feature information extraction,and estimation of relative position and attitude parameters of UAV has been deeply studied.The problem oftarget tracking and pose calculation based on vision is mainly discussed.On the target tracking problem,the "H" shaped landing target image is designed,the SIFT feature information is extracted and matched,and the matching result is further improved by RANSAC algorithm.Based on the advantages of Camshift tracking algorithm and particle filtering,a particle filter tracking algorithm combining target color feature,SIFT feature and texture feature is proposed,and it’s work process is described in detail.Firstly,the initial position and size of the tracking window are determined according to the SIFT matching results.Then,the Camshift algorithm is used to optimize the particle propagation,and adjust the size and direction of the tracking window adaptively,and finally estimate the state of the tracking target.At the same time,three fusion strategies of feature information are proposed,and particle resampling technology is introduced to deal with particle degradation.Experiments verify that the tracking algorithm is effective and feasible in the case of target occlusion and similar color interference.On the problem of pose calculation,the image of the tracked area captured by the airborne camera is subjected to image processing such as threshold segmentation,median filtering and corner detection.The image of the feature point of the "H" shaped landing sign is obtained on the image.Then,based on the transformation relationship between the perspective projection model and the coordinate system,the UAV poseestimation model is established.Finally,the singular value decomposition method are used to solve the optimal solution of position and attitude information.In the experiment,the position and attitude information of UAV itself is compared with the experimental pose information,and the error is within the allowable range,which verifies the correctness of the algorithm.更多还原