基于单目多视角影像的场景三维重建

3D Reconstruction of Scene Based on Monocular Multi-View Image

基于单目多视角影像完成场景的三维重建,首先从飞机航空影像中采用飞机及相机等各种信息提取出关键帧,然后对关键帧采用SIFT特征提取算法完成特征提取,提取出特征后采用比较最近邻与次近邻比值的方法完成特征点对匹配,为了提高匹配的鲁棒性,同时使用KVLD算法来完成误匹配点对的剔除。之后通过匹配点对计算出两两视图之间的旋转平移矩阵,从而进一步计算出全局坐标系下视图的位姿信息。根据视图的位姿信息通过三角化计算出匹配点对之间的三维稀疏点云,然后采用捆集优化策略来不算迭代优化,求出最佳的重投影矩阵,并优化稀疏点云。最后,在稀疏点云的基础上采用基于面片的三维多视角立体视觉算法进行稠密重建,并粘贴纹理得到可视化的场景三维点云模型。采用该方法能够仅根据多视角拍摄的图片快速、鲁棒地生成场景的三维重建模型,具有一定的应用价值。

In this paper,the 3D reconstruction of scene based on monocular multi view image,the key frames are extracted from the aerial images by using various information such as aircraft and camera,and then the key frames are extracted,the feature point pair are matched by using the ratio of nearest neighbor to next neighbor. At the same time,in order to improve the matching robustness,the KCLD algorithm is used to eliminate the mismatched point pairs. Then,the rotation transition matrix between the two views is calculated by matching point pairs,so as to further calculate the position and oriention information of the view in the global coordinate system. According to the position and orientation information of the view,the 3D sparse point between matching point pairs is calculated by triangulation,and then the optimal reprojection matrix obtained by using the bundling optimization strategy instead of iterative optimization,and the sparse point cloud is optimized. Finally,on the basis of the sparse point cloud,the 3D multi perspective stereo vision algorithm based on patch is used for dense reconstruction,and the texture is pasted to obtain visualization.