Registration Method for Point Cloud Based on Feature of One Plane and Two Cylindrical Holes

For improving the registration accuracy between point cloud and design model of the parts during scan detection, a registration method based on the feature of one plane and two cylindrical holes was proposed. The rough registration applies the features of plane and holes to coincide with the local coordinate systems of point cloud and design model. The fine registration is finished by using the improved iterative closest point (ICP) algorithm to calculate the sum of least squares of the distance between the closest points and point cloud. Since the calculation methods of registration areas and the closest points are different, fine registration is further classed into the global domain and feature domain types. The closest points of global registration are obtained by comparing the distance between point cloud and the nearest design model grid points or projection points, which is suitable for blank parts. The projection points of point cloud on a plane or a cylindrical hole are the closest points of feature domain registration, which is suitable for finished parts. The test results show that the accuracy of global registration will increase with the distance between surface discrete points decreasing. When the distance reaches 1.50 mm, the accuracy is about 0.15 mm, which meets the engineering requirements. For the same registration accuracy, the efficiency will be improved by 10%-20% compared with that of other methods.