基于PCA-SIFT特征融合的铸件冒口点云匹配研究

Research on Point Cloud Matching of Casting Risers Based on Fusion of PCA-SIFT Features

当前自动化铸件冒口切割机器人系统研制在识别和定位冒口方面要求精确,且面临现场环境复杂、相机采集数据量大、处理效率低、算法不稳定性导致误差较大等问题。针对上述问题,提出PCA-SIFTS4的点云配准方法,首先,对采集的场景点云数据通过多阶段滤波和聚类分割技术,从复杂的场景中提取出高质量的冒口点云;其次,基于PCA与3D-SIFT特征融合提取冒口点云的特征属性,降低Super4PCS粗配准的搜索复杂度,缩短配准时间;最后,采用点到面的ICP算法进行精配准,运用所提出的方法对典型形状冒口点云进行配准实验,与传统配准算法相比,配准时间和均方根误差分别平均降低78.53%和64.93%,结果表明PCA-SIFTS4算法对于环境复杂、数据量大、特征不明显的冒口点云具有较好的实时性和精确性。

The current development of automated casting riser cutting robot system requires precision in identifying and positioning the riser,and faces problems such as complex scene environment,large amount of data collected by the camera,low processing efficiency,and unstable algorithms leading to large errors.Aiming at the above problems,this paper proposes a PCA-SIFTS4 point cloud alignment method,firstly,the collected field point cloud data are filtered by multi-stage filtering and clustering segmentation techniques to extract high-quality riser point clouds from complex scenes;secondly,the feature attributes of riser point clouds are extracted based on the fusion of PCA and 3D-SIFT features,which reduces the searching complexity of the Super4PCS coarse alignment and Finally,the point-to-face ICP algorithm is used for fine alignment.In this paper,the proposed method is applied to the typical shape of the riser point cloud for the alignment experiment,compared with the traditional alignment algorithm,the alignment time and root mean square error are reduced by 78.53%and 64.93%on average,which shows that the PCA-SIFTS4 algorithm has better real-time and accuracy for the riser point cloud with complex environment,large data volume and inconspicuous features.