中文摘要：

基于面片的多视角立体（PMVS）算法只需要输入一组图片及摄像机参数便可生成稠密的带有颜色的三维点云,它无需任何的初始信息。但该算法仍有一些不足,一方面重建出的表面不够光滑和连续,特别是在大场景重建时经常碰到的俯仰拍摄情况尤为严重;另一方面对大型图像集进行重建时时间和空间复杂度高,无法实现重建。针对这些不足,提出了一种新的基于簇的PMVS改进算法。首先对图像集进行基于簇的划分,很大程度上减少了算法的运行时间,解决了系统内存消耗过大的问题,然后采用加入空间几何约束的PMVS改进算法对每个图像簇进行重建,提高了重建精度和表面光滑性,最后将所有簇的重建结果进行融合。通过实验证明了该算法的实用性和有效性。

英文摘要：

The patch-based multiview stereo（ PMVS） algorithm generates colorful 3D points only with a set of pictures and camera parameters,not needing any initial information. But there are two aspects that need to be improved. On the one hand,the reconstructed surface is neither smooth,nor continuous enough,and the problem becomes more serious under downward-shooting or upward-shooting,frequently accurred in large scene reconstruction. On the other hand,time and space complexity are too high to reconstruct a large image set. A novel PMVS algorithm based on clustering is proposed. It firstly decomposed the collection into a set of overlapping sets of photos based on clustering,so as to reduce the running time to a great extent and solve the problem that system memory consumption is too large; then reconstructed each cluster through the PMVS algorithm with geometric constraint to improve the accuracy and smoothness. The experiments show availability and practicability of the proposed method.