Ship Segmentation and Georeferencing from Static Oblique View Images
Borja Carrillo-Perez1, Sarah Barnes1, Maurice Stephan1
1German Aerospace Center (DLR), Institute for the Protection of Maritime Infrastructures, Fischkai 1, 27572 Bremerhaven, Germany.
This study introduces the ShipSG dataset for automated ship recognition and georeferencing in maritime surveillance. It evaluates segmentation methods, achieving high accuracy for real-time vessel monitoring and situational awareness.
Area of Science:
- Computer Vision
- Maritime Surveillance
- Geospatial Analysis
Background:
- Effective maritime situational awareness relies on accurate ship monitoring at ports.
- Current systems often lack the automation needed for real-time analysis and visualization of vessel data.
- Automated recognition and georeferencing of ships are crucial for enhancing maritime security and operational efficiency.
Purpose of the Study:
- To introduce the novel ShipSG dataset for ship segmentation and georeferencing in static oblique maritime monitoring scenes.
- To evaluate the performance of four instance segmentation methods (Mask-RCNN, DetectoRS, YOLACT, Centermask-Lite) for ship recognition.
- To propose and validate a novel method for georeferencing segmented ship masks using homography transformation.
Main Methods:
- Development of the ShipSG dataset featuring segmented and georeferenced ships.
- Comparative analysis of Mask-RCNN, DetectoRS, YOLACT, and Centermask-Lite for ship instance segmentation.
- Implementation of a homography-based method for transforming segmented ship pixels into geographic coordinates.
Main Results:
- DetectoRS achieved the highest mean Average Precision (mAP) of 0.747 for ship segmentation.
- Centermask-Lite demonstrated the fastest performance at 40.96 Frames Per Second (FPS).
- The proposed georeferencing method achieved accuracies of (22 ± 10) m within 400 m and (53 ± 24) m beyond 400 m.
Conclusions:
- The ShipSG dataset and proposed georeferencing method significantly advance automated maritime surveillance capabilities.
- Robust and real-time ship segmentation models are essential for effective maritime situational awareness.
- Accurate georeferencing of detected vessels enhances the utility of camera systems for port security and traffic management.
Related Concept Videos
Design Example: Identifying the Locations of Monuments in the Field Using Global Positioning System Device
Types of Global Positioning System Surveys
Geoid and Ellipsoid
Methods of Obtaining Topography
Field Application of Global Positioning System
Introduction to Surveying, Plane Surveying and Geodetic Surveys


