基于特征的点云精确配准算法

ACCURATE REGISTRATION ALGORITHM FOR POINT CLOUD BASED ON PROPERTIES

在散乱点云的配准过程中,由于不同次扫描得到的点云模型之间的重叠部分可能较小且点云具有丰富的几何细节,致使传统ICP(Iterative Closest Point)精确配准算法很难得到理想精度。针对这个问题以Chen和Medioni提出的点面距离误差测度函数为基础,结合基于特征的点云配准思想,设计了一种先建立拥有接近的主曲率的匹配点对集合,然后将二次拟合曲面间的平均距离作为误差测度进行迭代优化的精确配准算法。该算法在微小距离精确配准的应用环境下能提供相对于传统ICP算法更好的精度和更高的效率。

In process of scattered point cloud registration, since the overlapping portions of point cloud models derived from scanning in different times are ＂always quite small, plus the point cloud has abundant geometric details, this makes the ideal accuracy becomes difficult to be gained by traditional ICP accurate registration algorithm. In light of this problem, we design an accurate registration algorithm, it is based on the metric function of point to surface distance error put forward by Chen and Medioni, and combining the property-based point cloud registration idea. First it establishes matching points set with closed main curvatures, and then it takes the average distance between quadric fitting surfaces as the error metric for iterative optimisation. In application environment of minute distance, this algorithm can provide better precision and higher efficiency than the traditional ICP algorithm.