Automatic Coarse Registration of Urban Point Clouds Using Line-Planar Semantic Structural Features

Point cloud registration is essential for constructing comprehensive geometric information of a scene. In feature-based point cloud registration, the extraction of stable features and accurate feature matching significantly enhance the algorithm’s robustness and overall effectiveness. This article introduces an automatic coarse registration method for large-scale urban point clouds utilizing line-planar semantic structural features (LSSFs), specifically designed for structured urban environments. The proposed method begins by extracting plane and line features from the point cloud and constructing LSSFs based on their geometric relationships. Then, it establishes correspondences between the source and target point clouds utilizing a hash table and the four line-planar spatial information constraints (FLSICs). The method then groups potential semantic features through a four-point geometric consistency (FPGC) approach, from which it selects the most reliable feature group to estimate the initial spatial transformation parameters. The registration result is further optimized through a global maximum consensus set (GMCS), achieving accurate registration of the point clouds. Experiments conducted on six large-scale urban point cloud datasets demonstrate the effectiveness of the proposed method in achieving pairwise point cloud registration, with a rotation error of merely 0.07° and a translation error of 0.11 m. In comparison to existing methods, this approach significantly enhances registration efficiency and accuracy, particularly when dealing with high-density point clouds and in the presence of noise. The method offers a reliable and efficient point cloud registration solution for complex urban environments, indicating potential for widespread application.