<p>Accurate 3D pose estimation of overhead contact system (OCS) cantilevers is essential for collision-aware planning and safe robotic maintenance on railway vehicles. In practical deployment, however, strong stereo calibration is difficult to maintain due to large installation baselines, platform positioning errors, and long-term vibration and thermal drift. To address these challenges, this paper proposes a weakly calibrated dual-view 3D pose estimation framework for OCS cantilever structures, relying only on two monocular cameras and available engineering priors. The method exploits a frontal-oblique view and an upward-looking auxiliary view to provide complementary geometric constraints under weak extrinsic calibration. Cantilever pose recovery is formulated as a constrained nonlinear optimization problem that integrates dual-view reprojection consistency with structural priors, including fixed rod lengths and known assembly relations. An adaptive inter-view weighting strategy is further introduced to handle occlusion and unequal observation reliability between views, thereby improving robustness under practical sensing conditions. Comprehensive simulation studies evaluate the proposed method under increasing image noise, geometric degeneracy caused by feature overlap, and progressive mismatch of structural priors. The results reveal an accuracy–robustness trade-off in the proposed weakly calibrated formulation, which delivers competitive accuracy under nominal conditions and degrades gracefully while remaining stable in challenging scenarios where triangulation becomes ill-conditioned. A field experiment conducted on an in-service railway line further validates the practicality of the proposed framework, demonstrating that the reconstructed spatial envelope of the cantilever provides sufficient accuracy to meet on-site measurement requirements. These results confirm that the proposed approach provides a robust solution for vision-based cantilever pose estimation in real OCS maintenance environments.</p>