【摘要】 随着科技的发展，无损检测技术日趋成熟，对大型工件的检测结果进行三维可视化成像，并复现其整体空间形态具有重要意义。采用双目视觉系统与超声相控阵系统获取单次检测的三维可视化点云，对获取的超声点云采用3D局部特征描述算法提取具有代表性的点云特征点，之后使用快速点特征值直方图获取对应关系，并使用改进的随机抽样一致性算法提高映射关系的精度，以此对两块点云进行粗匹配，误差1.74 mm，采用最近点迭代算法精配准后误差仅0.13 mm，共用时16.5 s。之后对48块点云进行分阶段配准，最终误差为1.17 mm，能够满足实际检测中的准确性要求。更多还原

【Abstract】 With the development of science and technology,non-destructive testing technology is becoming more and more mature. It is of great significance to visualize the inspection results of large workpieces in 3D and reproduce their overall spatial shape. In this paper,a binocular vision system and an ultrasonic phased array system are used to obtain a single inspection. The 3D visual point cloud of the obtained ultrasound point cloud is extracted by the 3DScale Invariant Feature Transform algorithm to extract the representative point cloud feature points,and then the fast point eigenvalue histogram is used to obtain the corresponding relationship,and the improved random sampling consistency algorithm is used to improve the mapping relationship. The accuracy of the two point clouds is roughly matched,and the error is 1.74 mm. The error is only 0.13 mm after accurate registration using the closest point iteration algorithm,and the sharing time is 16.5 s. After that,48 point clouds are registered in stages,and the final error is 1.17 mm,which can meet the accuracy requirements in actual detection.更多还原