Multi-view point clouds registration method based on planar target

Multi-view point clouds registration is one of the key techniques in 3D vision measurement for large object. A planar target with some corners as feature points was designed. The distances between those feature points were accurately known. The target was placed at the common region which was measured by a vision sensor at two different view-points, and the vision sensor measured the 3D coordinates of the feature points twice. By establishing unit orthogonal basis using 3 arbitrary non-colinear points on the target, the original coordinate frame transformation matrix for multi-view point clouds was found. An objective function for the sum of distances' squares between the correspondent feature points was established after original registration, and the distances control was introduced to increase constraint. The optimal results were finally estimated by Levenberg-Marquardt optimization algorithm taking original coordinate frame transformation matrix as the initial value. A plaster model was measured by a binocular vision sensor at two positions. The experimental results show that the proposed registration method is simple and reliable, and the registration accuracy after optimizing is improved by about 31%.