自适应邻域半径选取的球形区域特征直方图特征提取算法

SRFH Feature Extraction Algorithm Based on Adaptive Neighborhood Selection

为提高点云特征描述符提取速度,通常使用球形区域特征直方图(SRFH)点云特征描述符提取方法,但此算法需要手动选取邻域半径,对参数的要求过高,若选取不当则大大降低配准的精度和速度。为此,提出了基于邻域半径自适应选取的SRFH特征提取算法。首先,计算多对点云的平均距离密度;设置多组邻域半径进行SRFH特征提取,然后配准,统计配准性能最优时的半径与点云密度;最后,通过三次样条插值拟合法求出最优函数,得到SRFH邻域半径自适应选取算法。实验结果表明,该算法根据点云密度自动选取最佳的邻域半径,点云配准速度快,配准精度高。

In order to improve the speed of point cloud feature descriptor extraction,the spherical region feature histogram(SRFH)point cloud feature descriptor extraction method is usually used.But the neighborhood radius needs to be manually selected.The requirements for parameters are too high,improper selection will greatly reduce the accuracy and speed of registration.An algorithm was proposed to adaptively select the neighborhood radius of the SRFH feature extraction.First,the average distance density of the multipair point cloud was calculated.Then,the radius and point cloud density with optimal registration performance were estimated after SRFH feature extraction and registration with setting multiple sets of the neighborhood radius.Finally,the optimal function was obtained using the cubic spline interpolation,the SRFH feature extraction algorithm based on adaptive neighborhood radius selection was developed.The experimental result shows that the algorithm which automatically select the optimal neighborhood radius according to the point cloud density improve the speed and accuracy of point cloud registration.