CAJ | 학술논문

以小型无人机对地观测为应用背景，研究了近似平面场景多视点图像拼接问题。对于已知粗略相机位姿的情况，提出一种融合相机位姿信息和图像特征点对应信息的方法，采用直接稀疏Cholesky分解方法求解拼接全局优化问题。由该方法得到的拼接结果没有全局变形，局部拼接误差也得到了明显的改善。对于相机位姿未知的情况，先采用structure-from-motion （SFM）方法恢复相机姿态和场景稀疏结构信息，再采用稀疏全局调整方法获得最终的图像变换参数。通过沙盘图像和真实的航拍图像拼接实验验证了算法的有效性。
이소형무인궤대지관측위응용배경，연구료근사평면장경다시점도상병접문제。대우이지조략상궤위자적정황，제출일충융합상궤위자신식화도상특정점대응신식적방법，채용직접희소Cholesky분해방법구해병접전국우화문제。유해방법득도적병접결과몰유전국변형，국부병접오차야득도료명현적개선。대우상궤위자미지적정황，선채용structure-from-motion （SFM）방법회복상궤자태화장경희소결구신식，재채용희소전국조정방법획득최종적도상변환삼수。통과사반도상화진실적항박도상병접실험험증료산법적유효성。
Aiming at applications of small unmanned aerial vehicle (UAV)in earth observations,this research explores the multi-viewpoint image mosaicing problems for roughly planar scenes.When coarse camera poses are known,a method for integrating camera poses and feature correspondences is proposed,in which direct sparse Cholesky factorization algorithm is used to solve the global optimization problems of mosaicing. Global distortions do not exist in the obtained mosaics and local mosaic errors are suppressed effectively.When the camera poses are unknown,a structure-from-motion (SFM)system is used to recover the cameras poses and sparse structure of the scene firstly.Then,the sparse global adjustment is used to refine the transformations.The proposed algorithm is validated using sand table images and real aerial images.