【摘要】 针对目前无人机图像处理自动化程度不高,限制条件较多,大数据量处理耗时较长的缺陷,本文提出一种基于单应性矩阵(Homography Matrix)的整体优化无人机图像拼接算法。本文采用GPU(图形处理器)+CPU协同工作模式,在CUDA架构下来实现SIFT算法,并以此算法对无人机影像提取特征点,采用极线约束和RANSAC算法剔除错点及外点,并估算出各张影像的变换矩阵Homography各参数的初始值,再采用LM算法精确求解出每张影像对应于基准面的全局Homography,实现全局配准,进而实现无人机影像的快速拼接。更多还原

【Abstract】 This paper concerns the problem of fully automated UAV(Unmanned Aircraft Vehicle) image stitching. Previous approaches have used human input or restrictions on the image sequence in order to establish matching images. In this paper, SIFT algorithm is achieved based on CUDA, and then we use this algorithm to detect feature points from the UAV images. After that we eliminate "Outliner-points " and error points by epipolar constraint and RASANC algorithm. At the same time, the initial value of every Homography matrix is estimated. In order to improve the accuracy of matching and registration, Levenberg-Marquet algorithm is used. During optimization, all control points are iterated separately to accelerate the processing and improve the accuracy of mosaic.更多还原