高反光物体的局部点云孔洞补全方法

Local point cloud hole completion method for highly reflective objects

针对高反光物体点云数据存在噪声明显、孔洞过大,使用传统贪婪投影三角化算法无法达到点云补全目的的问题,提出一种改进的贪婪三角化点云补全算法。利用统计滤波和高斯滤波进行离群点去除和平滑处理;采用移动最小二乘法对局部点云进行上采样增强数据,并且对点云进行进一步平滑。将贪婪三角化算法中的kd树(k-dimensional tree)算法替换为效率更高的八叉树(octree)搜索算法,将主成分分析法替换为准确率更高的移动最小二乘法进行法线估计;最后,进行点云三角化,完成点云补全。实验结果表明,改进算法能够更好地补全孔洞,补全后表面更加平滑、结构更为准确,且花费时间更短。

To solve the problem of the traditional greedy projection triangulation algorithm failing to complete point cloud due to significant noise and oversized holes in point cloud data of highly reflective objects,an improved greedy triangulation point cloud completion algorithm was proposed.Statistical filtering and Gaussian filtering were used for outlier removal and smoothing process,while the moving least squares algorithm was utilized to upsample and enhance the local point cloud data to further smooth the point cloud.The k-dimensional tree algorithm in the greedy triangulation algorithm was replaced by a more efficient octree search algorithm,and the principal component analysis was replaced by moving least squares algorithm with higher accuracy for normal estimation.Finally,point cloud triangulation was performed to complete point cloud completion.The experimental results show that the improved algorithm can better complete the holes,resulting in a smoother surface,more accurate structure,and less processing time.