Incorrect rebar placement has led to serious concrete structural failures; thus, site inspectors must ensure that the rebars placed are compliant with asdesigned drawings before tunnel lining concreting. This paper aims to develop a methodology that automatically segments the semantics (e.g., main rebar and distribution rebar) and instances of curved rebars and reconstructs the tunnel lining rebars from the raw point clouds. There are five main parts in the proposed methodology: rebar mesh reshaping, rebar mesh extraction and refinement, semantic segmentation, instance labeling, and rebar mesh modeling. To validate the developed methodology, an experiment was conducted on the rebar point cloud of a high-speed railway tunnel. The evaluation of the accuracies of segmentation was conducted by comparisons with the existing methods and manually labeled ground truth data. The results show that the rebar segmentation has high precision values with the acceptable recall values, and that the rebar models reconstructed achieve millimeter-level (about 2 mm) accuracy.