基于RBF神经网络的三维点云数据孔洞修补

Hole Repair of 3D Point Cloud Data Based on RBF Neural Network

三维扫描获取点云数据,往往由于被测物体自身形状复杂,扫描设备本身局限或者外部遮挡而出现孔洞,影响后续重构精度。由此提出一种基于RBF神经网络的三维扫描点云数据孔洞修补方法。该方法首先基于法线信息和KD tree提取三维点云的孔洞边缘信息,基于蒙特卡罗法在特征平面内生成填充数据点;然后将采集到的孔洞边缘特征点作为样本点集,训练RBF神经网络;最后,利用训练好的径向基函数,将二维特征平面内填充的数据点调整到孔洞区域的三维曲面,完成孔洞区域的数据修复。通过算例证实,该算法能有效的修补流形曲面的孔洞区域,并且修复区域和原有孔洞边界光滑连接,较好恢复点云模型原本的形貌结构。

When 3D scanning is used to obtain point cloud data,holes often appear due to the complex shape of the measured object itself,limitations of the scanning device itself,or external occlusion,which affect subsequent reconstruction accuracy.This paper proposes a 3D scanning point cloud data hole repair method based on RBF neural network.Firstly,based on the normal information and KD tree,the hole edge information of the 3D point cloud is obtained.Filled data points are generated in the feature plane based on the Monte Carlo method.Secondly,the collected feature points at the edge of the hole are used as a sample point set to train the RBF neural network.Finally,using the trained radial basis function,the data points filled in the two-dimensional feature plane are adjusted to the three-dimensional surface of the hole area,and the data restoration of the hole area is effectively completed.The experimental results show that the algorithm can well repair the hole area of the manifold surface,and the repair area can be smoothly connected with the original hole boundary,and the original morphological features of the point cloud model are well restored.