Unconstrained registration of 3-D scattered point clouds for free-form shape measurement

In combination of local surface fitting and generalized bisection optimization search,an automatic registration method is proposed for the multi-view 3-D scattered point cloud registration in the shape measurement of a large scale free-form surface.First,the standard least square surface is fitted in a small local area of point clouds.According to the fitting residue,all the non-flat regions for given point clouds are extracted.Based on the definition of "adjacency" and "reachability" of graph theory and the statistical characteristics of spatial distribution of non-flat regions,the non-flat regions for pairwise adjacent point clouds are clustered and matched coarsely.Thereby,the initial transformation is obtained.Then,a point in source point cloud is given and the most closed point and its corresponding k neighboring points in destination point cloud are calculated.Furthermore,the Moving Least Squares(MLS) surface fitting is performed for the k neighboring points.The given point and its orthogonal projection point on the MLS surface are selected as the matching points.Finally,the generalized bisection optimization search is employed to optimize the transformation.Experimental results indicate that the proposed registration method is stable,reliable and without human interaction.It is also suitable for the situation of offset sampling.The average registration seam is about 0.02 mm when 150 matching points in overlapping region are used in optimization calculation.The proposed method meets the requirements of multi-view 3-D scattered point cloud registration in the shape measurement of large scale free-form surfaces.