Vision Control for Cable Binding Robot in Offshore and Marine Industry

Abstract

The development of a robotic system aimed at enhancing the cable binding process for Floating Production Storage Offloading (FPSO) vessels addresses critical challenges within the labour-intensive shipbuilding industry. The proposed robotic system incorporates computer vision technology which has the ability to detect stainless steel buckles and relay their precise locations to improve the efficiency and effectiveness of the overall process. This study divides the entire process into three steps: object detection, key point prediction, and pose estimation. We have leveraged You Only Look Once (YOLOv8) to detect and predict the key points of the stainless steel buckles. The study has successfully tackled a significant challenge in machine vision, namely detecting the shiny metal parts of the buckle against a similarly shiny background. Furthermore, we validated the effectiveness of our computer vision approach by adopting a training approach that combines real and synthetic data, which mitigates overfitting issues during neural network training.