一种改进法向量估算的点云特征提取

Feature extraction from point cloud based on improved normal vector

针对传统主成分分析点云法向量估算存在精度不高、人工干预等问题,提出了一种改进的主成分分析法向量估算方法,并根据点邻域法向量标准差实现了离散点云特征提取,最后对关键参数K取值进行讨论。通过引入高斯核函数为邻域点集权值计算公式,削弱了距离较远点对采样点法向量精度影响;以信息熵最小化为约束条件自适应选择最佳邻域尺度R,较好的顾及了局部点云空间特征的差异性。实验结果表明:该方法能很好的实现点云特征提取,改进后的点云法向量估算顾及点云邻域复杂程度,具有普适性强、自动化程度高的特点,建议关键参数K取值范围12~20。研究对点云配准、数据精简及模型重构等点云预处理研究具有理论参考价值。

In view of the shortcomings of traditional principal components analysis(PCA)point cloud normal vector estimation,such as low accuracy and manual intervention,an improved PCA normal vector estimation method is proposed,and the feature extraction of discrete point cloud is realized according to the standard deviation of point neighborhood normal vector.Finally,the value of key parameter k is discussed.Gauss kernel function is introduced to calculate the weight value of neighborhood points,which weakens the influence of distance points on the accuracy of normal vector of sampling points;Taking the minimization of information entropy as the constraint condition,the optimal neighborhood scale R is adaptively selected,which takes into account the difference of spatial characteristics of local point clouds.The experimental results show that this method can achieve the point cloud feature extraction well,the improved point cloud normal vector estimation takes into account the complexity of the point cloud neighborhood,and has the characteristics of strong universality and high degree of automation.It is suggested that the value range of key parameter Kis 12~20.The research has theoretical reference value for point cloud registration,data reduction and model reconstruction.