<p>The fine registration of ground point clouds is crucial for the application of planar tunnel model reconstruction and lining structure identification, but the presence of many repetitive structural and weak geometric features in planar tunnels, coupled with scanning obscuration poses a great challenge to the registration. Aiming at the difficulty of pairwise fine registration of point clouds in planar tunnel scenarios, a registration strategy under various restrictive conditions is proposed in this paper. The transformation matrix with six parameters is solved in three parts. The bottom surface is used to solve for two small rotation angles and height differences, and the number of point clouds is reduced by extracting the side point clouds from the bottom-corrected point clouds, which increases the computational efficiency. The key points are extracted to improve the quality by normal vector differences and are classified into two categories, where only the key points on the same side will produce correspondences. Finding key points where geometric features are most similar under increasingly stringent distance constraints. The Gauss–Markov (G–M) model generated by dimensionality reduction and refinement of the Bursa 7-parameter transformation model was used to solve for the remaining three parameters. Experimental evaluation of point cloud data acquired in real tunnels shows that the proposed method performs well in terms of accuracy, robustness, and runtime.</p>