Abstract: Point clouds reduction is a key step before the triangulation in the progress of 3D-reconstruction which based on images. Generally, the quantity and distributed mass of the point clouds after reducing will contribute to the result of reconstruction directly. This paper keeps the point clouds’ character that the point clouds after reducing by the traditional method distribute adaptively according to curvatures. Considering the disadvantages of traditional method that holes will be found in the flat areas and balanced distribution won’t be guaranteed. The paper gives an improved algorithm to solve the problem. Firstly, it uses bounding box with small grid to simplify the point clouds which will also make it easy for the calculation of point’s local characters. Then it classifies the point clouds and respectively simplifies them based on their normal angle coefficients and the degrees of bending. In the end, the paper gives an easy method to evaluate the result of point clouds after reducing. It has been proved that, the method of this paper can keep the geometrical features well, and can overcome the shortcomings of the traditional method, too.

Key words: 3D-reconstruction, point clouds reduction, classification and respective sampling, feature preserving

摘要： 点云精简是基于图像的三维重建过程中的一个关键步骤，精简后点云的数量和分布质量将直接决定重建的效率。针对传统方法易在平缓区出现孔洞和无法保证均衡分布的缺点，在保留传统精简方法精简后点云依据曲率自适应分布特点的基础上，给出了一种改进的点云分类精简算法。首先对点云进行小栅格包围盒精简，可初步简化点云并方便点云的特征计算，然后通过点的法向夹角系数和弯曲度进行点云分类并分别采样，最后给出了一种简易的方法来进行精简效果的评估。实验结果表明，该方法能较好保留点云几何特征，克服传统方法的缺点。

关键词: 三维重建, 点云精简, 分类采样, 特征保留