The growing need for efficiency and accuracy in shipbuilding has driven the adoption of advanced automation and perception systems, particularly for repetitive and precision-dependent tasks. This work presents a hybrid computer vision approach aimed at improving the identification of small pre-assembly components within a robotic welding cell. The proposed system integrates object detection using deep learning with 3D surface and edge matching, allowing for more efficient processing compared to conventional point cloud-only methods. A dataset of real images was used to train and evaluate two state-of-the-art detection models, with the best performing model selected for integration into the system. By limiting 3D analysis to the regions identified through 2D detection, the proposed approach reduces computational overhead while improving recognition speed and robustness, offering a more reliable and scalable solution for industrial environments.