An End-to-End Platform for Managing Third-Party Risks in Oil Pipelines

Abstract

Ensuring the safe and reliable operation of underground oil pipelines is crucial to prevent environmental disasters and maintain uninterrupted energy supply. Yet, this vast network faces threats from third-party activities, natural disasters, and aging infrastructure, posing risks of catastrophic consequences if left unaddressed. In response to this need, this paper presents a computer vision system for detecting third-party risks (vehicular movement) near oil pipelines. Our primary objective is to showcase the practical application of cutting-edge computer vision models in real-world operational environments. For this, we construct a dataset comprising 1,003 aerial images, covering seven classes of vehicles commonly encountered near pipelines, including trucks, forklifts, machinery, pickups, tractors, vehicles, and buses. This dataset serves as the foundation for training and hyperparameter optimization of a YOLOv8x-based detection model, used in this work. The optimized model exhibits strong performance across precision, recall, F1-score, and mean average precision metrics compared to the baseline model. Additionally, graphical tests illustrated that the optimized model demonstrates higher confidence scores and a reduction in false positives. In addition, a platform has been developed to seamlessly integrate the model. This platform offers a range of functionalities, enabling users to access the alert history, prioritize alerts, track actions taken on each alert, visualize alerts geographically, receive notifications for identified risks, and generate detailed reports for comprehensive analysis and decision-making.