一种结合纹理择优算法的影像三维重建方法

发布时间：2018-06-04 15:28

  本文选题：纹理映射 + 三维重建　； 参考：《测绘科学》2017年01期

【摘要】：针对多视影像重叠度高、影像来源丰富等特点,提出了一种基于多视角影像的纹理择优映射算法,该方法对模型三角形进行逐个相机场景可见性分析,对模型在场景中不可见三角形和部分可见三角形进行选择性剔除,只对完全可见三角形及符合阈值计算的部分可见三角形提供候选纹理三角形,能有效解决模型不可见三角形和部分可见三角形被误贴纹理的问题,再通过对候选纹理三角形的视角分析,为几何模型表面三角形选择一个理论最优纹理,计算映射关系自动映射到模型表面。同时,本文将该纹理择优映射算法应用到基于近景影像的三维重建中,使用从运动中恢复结构SFM的方法进行相机标定及影像相对定向,通过CMVS/PMVS密集匹配方法从影像中获取点云模型,采用Possion算法重构模型三角网,最终利用提出的纹理择优算法确定最佳纹理并实现自动映射。通过与Smart3d、PhotoScan、lensphoto软件的对比证明了本文三维重建及纹理择优算法在近景影像三维重建中的有效性。
[Abstract]:In view of the characteristics of high overlap and abundant sources of multi-view images, a texture optimal mapping algorithm based on multi-view images is proposed, which analyzes the visibility of model triangles on camera scenes. The model is selectively removed from invisible triangles and partially visible triangles in the scene, and only candidate texture triangles are provided for fully visible triangles and partially visible triangles that conform to the threshold calculation. It can effectively solve the problem of invisible triangles and partially visible triangles being misaffixed to textures, and then through analyzing the angle of view of candidate texture triangles, we can select a theoretical optimal texture for geometric model surface triangles. The mapping relation is automatically mapped to the surface of the model. At the same time, the texture optimal mapping algorithm is applied to 3D reconstruction based on close-range image, and the method of recovering structure SFM from motion is used for camera calibration and image orientation. The point cloud model is obtained from the image by CMVS/PMVS dense matching method, and the model triangulation is reconstructed by using the Possion algorithm. Finally, the proposed texture selection algorithm is used to determine the best texture and realize the automatic mapping. The comparison with Smart3dX PhotoScanlen-sphoto software proves the effectiveness of the proposed 3D reconstruction algorithm and texture selection algorithm in the 3D reconstruction of close-range images.
【作者单位】： 长沙市规划信息服务中心;中南大学地球科学与信息物理学院;
【基金】：国家“973”计划资助项目(2012CB719904)
【分类号】：TP391.41
，

本文编号：1977881