【作者】 李申；

【导师】 李忠海；

【作者基本信息】 沈阳航空航天大学 ， 模式识别与智能系统， 2012， 硕士

【摘要】 为了提高无人机着陆自主性和智能性,本文针对固定翼无人机研究了基于视觉辅助的自主着陆方法。通过对无人机着陆过程的分析,将无人机利用视觉着陆的过程分为三个阶段,即远距离阶段、中距离阶段和近距离阶段。在每一阶段采用不同的视觉策略,识别不同的目标,达到不同的识别精度。在远距离阶段,无人机视觉识别的主要任务是通过对机场及其周围主要环境标识物的识别来确定机场的位置。对机场的识别主要是将待识别图像分割成16*16大小的子块,通过Contoulet变换对子块进行多尺度分解,在不同尺度下提取图像的矩特征和纹理特征,并将这些特征作为该子块的特征向量。将每个子块的特征向量与建立的机场环境标识的特征库相对比来识别该子块属于的环境类别,最终达到识别机场的目的。在中距离阶段,无人机识别的主要目标是跑道。对于跑道识别本文主要采用模板匹配的方法。利用改进的快速SIFT算法提取模板和背景的SIFT特征,并将模板的SIFT特征与背景中的所有SIFT特征相对比,寻找与模板匹配的特征点,当背景中存在一定量的匹配特征点时,就认为背景中有跑道,并确定跑道的具体位置。在近距离阶段,无人机视觉识别的主要任务是识别着陆标识。本文设计“H”型的着陆标识,并利用着陆标识与跑道颜色的差异提取出着陆标识。由于实际环境的复杂性,本文首先采用稳健的Hough变换检测标识边缘直线,然后通过求取直线方程交点的方法来提取着陆标识上的特征点。利用平面的单应性对摄像机进行在线标定,通过建立坐标系和解算摄像机的外参数矩阵估计无人机的姿态和位置。将无人机视觉着陆过程划分为三个阶段具有一定的科学性,每个阶段之间有一定的连续性,每个阶段的识别重点不同且识别任务较轻。实验结果表明,在远距离阶段的算法能够很好的识别地貌的主要环境标识,中距离的SIFT特征匹配能够很好的定位跑道的位置,近距离的在线标定和姿态估计技术能够满足无人的着陆需要。更多还原

【Abstract】 In order to increase autonomy and intellectuality of the UAV landing, this paper studies the visual assistant landing method of fixed wings UAV. Through the analysis of UAV landing process, this paper divided the UAV landing process into three visual stages, which is remote stage, middle stage and close stage. In different stages, UAV uses different visual strategies, identify different goals, achieve different accuracies.In the remote stage, the main mission of the visual identification is confirming the orientation of the airport by identifying the airport and the main environmental objectives around it. The method of identifying airports is to divide the image into 16*16 sized subblocks firstly, making multi-scale decomposition of every subblock through the Contoulet transform, then extracting moment feature and texture feature under different scales, and combine these features as subblock’s feature vector. Comparing each subblock’s feature vector with the airport environmental feature database in order to identify which environment the subblock belongs to, finally determine the location of the airport in the view. In the middle stage, the runway is the main target that UAV need to identify from background. The template matching method can be used for the runway identification in this paper. The rapid improved SIFT algorithm is used to extract SIFT characteristics of the background and the template, then comparing the template’s SIFT features with the background’s, in order to look for the characteristic points matched with template in the background. There must be a runway in the background, when there are some matched feature points in the background, then getting the location of the runway. In the close stage, the main task of UAV visual recognition is identifying the landing logo. This paper designs landing logo as "H" and extracts it from background by using the difference of bright color between the landing logo and the runway. Because of the complexity of the actual environment, this paper uses the steady Hough transformation to detect line of logo’s edge and extract characteristic points on the logo by getting cross points of straight-line equation. Then making camera on-line calibration by using the homography of plane, getting attitude angle and location by establishing coordinate system and calculating external parameters matrix of the camera.There is scientificity in dividing UAV visual landing process into three stages, because each stage has certain continuity, the identifying key is different and recognition task is light in each stage. The experimental results show that the main environmental logo can be recognized well in the remote stage, the SIFT feature matching can locate the runway very well in the middle stage, the technology of on-line calibration and pose estimation can meet the need of the UAV landing in the close stage.更多还原