复杂零件点云的高精度鲁棒配准方法

High Precision Robust Registration Method for Point Clouds of Complex Parts

点云配准是三维测量中的关键一步,但由于零件点云表面相似特征多,误匹配概率大,导致配准结果难以保证。为此,提出了一种具有高鲁棒性、高精度的点云配准方法。首先,使用FPFH特征描述子来计算点云特征向量,产生初始匹配点对集。然后,依据具有旋转平移不变性的精确几何结构特征对初始匹配点对集进行筛选,剔除误匹配点对。最后,利用列文伯格-马夸尔特(L-M)算法计算点云之间的变换矩阵。实验结果表明,与其他方法相比,其配准精度评价指标RMSE降低80%以上,结合精配准方法可进一步将RMSE值降低86%,从结果可看出本文方法配准精度高且具有较高的鲁棒性。

Point cloud registration is a key step in 3D measurement.However,due to the similar features on the surface of the part point cloud,the probability of mismatching is high,which makes the registration result difficult to ensure high accuracy.A point cloud registration method with high robustness and high accuracy was proposed.Firstly,the FPFH descriptor was used to calculate the feature vector of the point cloud to generate the initial matches.Then,according to the precise geometric features with rotation and translation invariance,the initial matches was filtered,and the mismatched points were eliminated.Finally,the transformation between point clouds was calculated by Levenberg Marquardt(L-M)algorithm.The experimental results showed that,compared with other methods,the registration accuracy evaluation RMSE was reduced by more than 80%.Combined with the fine registration method,the RMSE value could be further reduced by 86%.From the results,it could be seen that this method had high registration accuracy and high robustness.