Point Cloud Registration Method Based On Heterogeneous Moving Least Square Surface

3D laser scanner is commonly used to measure mechanical workpiece and obtain data points set in reverse engineering.Data point set is obtained by multi-view scanning.Data points from different angles are defined in different coordinate systems.Therefore,it is necessary to register the multi-view scanning data to describe the overall shape of the workpiece.In engineering,the commonly used registration methods are hardware-based registration method and iterative closest point(ICP)based registration algorithm.The point-to-point ICP algorithm is simple and easy to implement and widely used.However,its registration accuracy is sensitive to sampling density and noise.Therefore,this paper proposes a multi-view cloud registration method based on heterogeneous moving least squares surface,and the research work is carried out in the following aspects:Geometric features of complex workpiece are rich and changeable.In order to achieve accurate registration,it is necessary to conduct global and local analysis of shape change characteristics over multi-view scanning data.Through analysis,geometric feature parameters describing shape change are extracted.In order to represent the change of heterogeneous shape feature,the principal curvature is used to characterize the change trend of any point in different directions.In order to estimate the principal curvature accurately and steadily,integral invariant analysis need to be carried out on a suitable scale.the scales should be consistent with local features and neighborhoods,which can be obtained iteratively by solving the principal curvature.Through the analysis of principal curvature,the description which is suitable for the change of global shape and local heterogeneous shape characteristics is obtained.A linear heterogeneous Gauss kernel function is constructed with principal curvature.The heterogeneous Gauss kernels is used to define heterogeneous moving least squares(HMLS)surface,which can depict local heterogeneous shape feature in the local neighborhood of the scanning data.HMLS surface can not only describe the shape feature of flat area,but also depict the sharp area shape feature.It can also reflect the change of local shape feature along two different directions,and can accurately approximate the geometric model of workpiece.On the basis of constructing HMLS surface in local neighborhood,the scanning data are projected onto HMLS surface,and the projection points are calculated.An orthogonal projection algorithm from point to HMLS surface is proposed,and the second-order convergence of the projection process is proved.Points are projected onto HMLS surface to construct corresponding point pairs.The rigid transformation matrix is calculated according to the point pairs,and the scanning data are registered to the HMLS surface.Finally,through simulation and practical engineering cases,the proposed method is proved to be robust to noise and can achieve accurate registration of multi-view clouds.