Structural steel members may experience nonlinear geometric instabilities, which can compromise structural safety. Reliable identification of such distortions and their effects is essential for post-event assessment and infrastructure maintenance. This paper proposes novel methodological developments for cross-sectional profile recognition and 3-dimensional (3D) geometric reconstruction of I-shaped steel beam-columns from 3D point clouds. The methodology applies to both undeformed and deformed members exhibiting inelastic local buckling. The proposed scan-to-model workflow automatically recognizes cross-sectional properties and centerline features from multiple section cuts along the member height, which are integrated to reconstruct the full 3D geometry while preserving key structural constraints. To support practical deployment, scanning noise is quantified under different environmental conditions and scanning devices. Guidance is provided on balancing scanning quality and acquisition time. The methodology is validated using physics-based simulated point clouds with augmented realistic noise as well as in-situ scans of 3D-printed and steel members with diverse geometric distortions. A near-100% cross-section recognition success rate is achieved. It is shown that compared to existing 3D reconstruction methods, the proposed methodology attains higher accuracy (mean error ∼ 2 mm), improved robustness, and higher computational efficiency, demonstrating its effectiveness for automated inspection.