Abstract: In the coarse registration of heterogeneous point clouds in architectural scenes,challenges such as large data volume,feature confusion,and severe interference from outliers hinder registration accuracy.To address these issues,this paper proposes a lightweight coarse registration algorithm based on building contours.First,the algorithm accurately extracts key contour points of buildings by leveraging the eigenvalue ratio of the point cloud normal distribution matrix.Then,it replaces traditional normals with the principal direction of points to compute fast point feature histograms(FPFH),and filters out incorrect correspondences using bidirectional consistency and geometric consistency constraints.Finally,the Geman-McClure function is used to construct a robust objective function to achieve accurate estimation of the transformation matrix.Experimental results demonstrate that the proposed algorithm outperforms existing methods in terms of registration accuracy and overlap ratio,thereby validating its effectiveness and reliability in coarse registration of heterogeneous point clouds in architectural scenes.

Key words: heterogeneous point cloud fusion, point cloud registration, contour keypoints, feature matching