Scholarlink

For over 30 years, remotely operated vehicles (ROVs) have monitored subsea operations in the offshore oil and gas industry. Recent advancements in ROV electronic devices and computational systems offer the potential to enhance data processing capabilities. This study focuses on improving subsea pipeline maintenance by integrating ROVs with advanced computational techniques, bridging the gap between ROVs and fully autonomous underwater vehicles. The methodology includes data collection, pre-processing, model training, evaluation, testing, and trajectory processing with the dead-reckoning algorithm. Pilot projects collect data, which is manually labeled and trained using predefined architectures. Evaluation metrics gauge model performance, and successful models enable trajectory reconstruction. Comparing ResNet18 and YOLOv4-tiny models for object detection, YOLOv4-tiny demonstrated superior accuracy with longer training times. The study also explores the impact of predicted angles on trajectory reconstruction, emphasizing the importance of selecting models tailored to specific application requirements.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Image-Based Underwater Pipeline Tracking Using Oriented Object Detection and Dead-Reckoning Algorithm

  • 章节

  • 发表日期：

    2026年

查看全文
  • Ahmad Lazuardi Iman,
  • Aulia Siti Aisjah,
  • Wasis Dwi Aryawan

摘要

For over 30 years, remotely operated vehicles (ROVs) have monitored subsea operations in the offshore oil and gas industry. Recent advancements in ROV electronic devices and computational systems offer the potential to enhance data processing capabilities. This study focuses on improving subsea pipeline maintenance by integrating ROVs with advanced computational techniques, bridging the gap between ROVs and fully autonomous underwater vehicles. The methodology includes data collection, pre-processing, model training, evaluation, testing, and trajectory processing with the dead-reckoning algorithm. Pilot projects collect data, which is manually labeled and trained using predefined architectures. Evaluation metrics gauge model performance, and successful models enable trajectory reconstruction. Comparing ResNet18 and YOLOv4-tiny models for object detection, YOLOv4-tiny demonstrated superior accuracy with longer training times. The study also explores the impact of predicted angles on trajectory reconstruction, emphasizing the importance of selecting models tailored to specific application requirements.