基于有结构信息参与的点对一致性投票的点云法向估计算法

Point Cloud Normal Estimation Algorithm Based on Point Pair Consistent Voting with Structural Information Participation

摘要: 基于平面拟合的算法被广泛地应用于估计点云的法向，近几年来，我们采用最小二乘拟合和加权的方法相结合的思想去估计点云的法向，但是对于尖锐特征附近的点，由于缺少点与平面的结构信息：点与平面法向之间的差异，就会使得法向异常的点对我们的影响比较大。为了减少这类点的影响，我们把邻域分割与平面拟合的思想相结合，提出了有结构信息参与的点对一致性投票的点云法向估计算法，它能更好地估计尖锐特征处点的法向。我们首先，对于当前点所使用的邻域构造协方差矩阵，刻画出点与尖锐特征的距离远近程度，然后再加入了点与平面的结构信息，最后使得拟合出来的正确平面评分较高。实验结果表明，我们的方法与PCV相比，在运行速度上基本一致。但是，我们对于点云法向的估计更为准确，对细节刻画地更好。

Abstract: The algorithm based on plane fitting is widely used to estimate the normal direction of point cloud. In recent years, we have adopted the idea of combining the least square fitting and the weighted method to estimate the normal direction of point cloud. However, for the points near sharp features, due to the lack of structural information of the point and plane: the difference between the point and the plane normal direction, the point with abnormal normal direction will have a greater impact on us. In order to reduce the influence of these points, we combine the idea of neighborhood segmentation and plane fitting, and propose a point cloud normal estimation algorithm with the participation of structure information, which can better estimate the normal direction of points with sharp features. First, we construct a covariance matrix for the neighborhood used by the current point, describe the distance between the point and the sharp feature, and then add the structure information of the point and the plane, and finally make the correct plane fit out have a large score. The experimental results show that our method is consistent with PCV in speed. However, our estimate of the normal direction of the point cloud is more accurate and better characterizes the details.