【作者】 李承光；

【导师】 徐锦法；

【作者基本信息】 南京航空航天大学 ， 航空宇航科学与技术， 2009， 硕士

【摘要】 无人直升机的自主着陆是其飞行过程中的重要阶段,对导航与控制能力都提出了很高的要求。目前常用的INS/GPS导航已不能满足其精度要求,利用计算机视觉完成该任务就成为当前的发展趋势。本文对目前国内外在自主着陆方面采用的各种计算机视觉技术进行了综述,并分析了各自的优缺点。本文在着陆场景为一平面假设的基础上,建立了无人直升机和平面着陆场景间相对运动的数学模型,以及摄像机的透视成像模型,以此为依据提出了一种无人直升机的运动估计方法:利用单目机载摄像机对着陆场景拍摄的图像序列,根据摄像机和着陆场景的相对运动模型和运动投影关系,由像平面上的2D运动恢复出直升机和摄像机的3D运动。像平面上的2D运动可通过计算连续图像帧间的光流得到。针对光流计算常用Horn-Schunk方法和Lucas-Kanade方法在计算较大像素运动时的不足之处,本文引入了基于图像金字塔结构的多分辨率、层次化的L-K迭代计算方法,对估计速度较大的运动能取得较好的效果。本文以Vega Prime视景仿真环境为平台,模拟直升机、着陆场景以及它们之间的相对运动,运用光流算法得到对运动参数的估计,并分析了图像频谱、直升机高度、摄像机视场等因素对算法的影响。在此基础上,对该算法的实时性、准确性、可靠性进行验证,指出了算法和实验设计的不足之处,最后,提出了一些改进意见并对下一步的工作做了展望。更多还原

【Abstract】 The autolanding of unmanned helicopters is an important stage of their flying process.it requires good capability of navigation and control.Now the common used INS/GPS navigation can’t satisfy its requirement of accuracy.The application of computer vision has become a development trend.This paper summarizes the details of a variety of computer vision technologies which are researched and applied in the autolanding of unmanned helicopters at home and abroad.On the basis of hypothesis that the autolanding scene is a plane, the mathematical model of relative motion between helicopter and the plane scene is built, also the perspective imaging model is built under the condition of same hypothesis. According to this, a motion estimation method is proposed.The image photos of landing scenarios is gotten with an onboard camera.Utilizing the relative motion and projection relation between the camera and landing scene, the 3D motion of the helicopter and camera is recovered from the 2D motion in the imaging plane.The 2D motion in the imaging plane can be obtained by calculating the optical flow between two successive image frames.The common methods used in computing optical flow such as Horn-Schunk method and Lucas-Kanade methd have much limitation in computing big pixel motion, so we introduce a multi-resolution, hierarchical L-K iteration method based on the pyramid structure of images.Thus good results can be achieved in estimation of high speed motion.In this paper, Vega visual simulation environment is adopted as experiment platform,which is used to simulate the helicopter,autolanding scene and the relative motion between them.The motion parameters are estimated applying the algorithm based on optical flow.The influence of some effecting factors to this algorithm is analysized, including image’s frequency, helicopter’s height, camera’s field of view, etc.On this basis, the correctness and reliability of this algorithm are verified, the lack and shortcoming in the algotithm and experiment design is also pointed out.At last, some improvement suggestions will be proposed, and we will make a prospect for future work.更多还原