飞机蒙皮表面平缓、光滑、特征少,在使用迭代最近点进行数模配准时会产生错位、局部最小值等问题,因此提出了一种基于轮廓约束的蒙皮点云配准方法。首先定义一种新的三维轮廓Cκ特征点描述方法,并基于距离约束对初始Cκ特征点进行聚类和过滤,实现对蒙皮点云特征的精确描述。其次,基于距离的快速点特征直方图(FPFH-d)特征相似度约束,寻找点云和数模特征点的对应点对,实现蒙皮轮廓的初始配准。最终基于迭代最近点算法,融合Cκ特征描述的轮廓约束,实现蒙皮的精配准。利用斯坦福公共数据库点云对算法的速度和精度进行测试,与点快速直方图算法(FPFH-SAC-IA)相比,初始配准精度分别提高了48.47%和77.29%,速度分别提高了70.35%和97.08%,证明了Cκ特征点提取算法的普适性和有效性。基于以上算法,对飞机蒙皮测量数据进行实验验证,配准准确度达到了100%,在12 m 3范围内全局误差优于3.5 mm。该方法有效解决了蒙皮配准时的错位和局部最小值问题。

When the iterative closest point (ICP) is used for model registration of the aircraft skin surface, problems such as dislocation and local minimum value will appear since the surface is flat and smooth and has few features. For this reason, a registration method of skin point clouds based on contour constraints was proposed. Firstly, a new description method of Cκ feature points for a three-dimensional contour was defined, and the initial Cκ feature points were clustered and filtered based on the distance constraint, realizing the accurate feature description of skin point clouds. Secondly, on the basis of the similarity constraint of fast point feature histogram based on distance (FPFH-d) features, the corresponding point pairs of point clouds and model feature points were found to achieve the initial registration of the skin contour. Finally, according to the ICP algorithm, the contour constraints of Cκ feature description were fused for the precise registration of the skin. Furthermore, the speed and accuracy of the new algorithm were tested by the point clouds from the Stanford public database. In comparison with the fast point feature histograms-sample consensus initial alignment (FPFH-SAC-IA), the initial registration accuracy of the proposed algorithm is improved by 48.47% and 77.29%, respectively, and the speed is increased by 70.35% and 97.08%, respectively, which proves the universality and effectiveness of the extraction algorithm of Cκ feature points. In addition, based on the proposed algorithm, we conduct experiments to verify the measurement data of aircraft skin and the results demonstrate that the registration accuracy reaches 100% and the global error is better than 3.5 mm in the range of 12 m ³. In conclusion, the method proposed in this paper can effectively solve the problems of dislocation and local minimum value during skin registration.