Detection-Based Object Tracking Applied to Remote Ship Inspection

Area of Science:

• Computer Vision
• Artificial Intelligence
• Maritime Engineering

Background:

• Manual ship inspections are labor-intensive and can be prone to human error.
• Automating the detection of structural defects like cracks is crucial for maritime safety and maintenance.

Purpose of the Study:

• To develop an automated detection and tracking system for identifying cracks in maritime ship inspection videos.
• To enhance the efficiency and accuracy of remote ship inspections using deep learning.

Main Methods:

• A two-stage system combining object detection (RetinaNet) and an enhanced tracking system (CSRT with data association).
• Customized RetinaNet with optimal anchor settings for crack detection.
• Improved CSRT tracker to mitigate model drift and a data association algorithm considering IoU and area for robust tracking.

Main Results:

• The proposed system demonstrated reasonable performance in automatically analyzing ship inspection videos.
• Successfully compensated for detection jitters, reducing tracking jitter and redundant trackers.
• Validated the feasibility of using deep neural network-based computer vision for automated remote ship inspection.

Conclusions:

• The developed detection-based tracking system offers a viable solution for automating maritime ship inspections.
• The system shows promise for integration into digital infrastructure to streamline the inspection process.
• Highlights the potential of AI and computer vision in enhancing maritime safety and operational efficiency.