TL;DR
This paper introduces a novel computer vision system for estimating the towing line angle in real-time during maritime towing operations. By leveraging image analysis of the sea-sky line for tug roll estimation and a specialized algorithm for towing line detection, the system provides crucial data for ensuring the safety and stability of these complex operations.
What Problem Does This Paper Address?
The increasing scale of offshore projects necessitates more frequent and complex towing operations of non-powered facilities. These operations are inherently risky due to the unpredictable nature of the marine environment (wind, waves, tides) and the complex hydrodynamics of large towed objects. A critical aspect of safe towing is monitoring the towing line angle, as excessive angles can lead to increased stress on the towline, potential breakage, and system instability. Existing methods often rely on expensive or complex sensor setups. This paper addresses the need for a cost-effective, real-time, and robust solution for estimating this crucial towing parameter using only camera vision.
Methods
The core of the proposed method lies in a two-pronged computer vision approach:
- Sea-Sky Line Detection for Roll Estimation: The system first identifies the sea-sky line in the captured images. This is achieved by analyzing vertical gradients, smoothing the image with a total variation algorithm to preserve edges, and then using the Random Sample Consensus (RANSAC) algorithm to robustly estimate the sea-sky line angle. This angle directly corresponds to the rolling angle of the towing vessel, which is a necessary input for accurate 3D world coordinate calculations.
- Towing Line Angle Estimation: A novel saliency detection method is employed to identify the towing line. This method analyzes brightness distribution and uses an adaptive threshold to select potential towing line points. Crucially, it incorporates specialized filters to mitigate noise from sea waves and stern spray, which are common challenges in maritime imagery. After filtering, the towing line angle in the 2D image plane is estimated using a least-squares method. This 2D angle, combined with the estimated rolling angle and camera geometry, is then used to calculate the towing line angle in the 3D world.
The geometrical projection model from 3D to 2D is detailed, and the camera is assumed to be mounted such that its optical axis is parallel to the sea surface and aligned with the vessel's longitudinal symmetry plane.
Datasets
The paper describes experimental validation using video data recorded during a real-world towing operation. Specifically, the vessel DONG HAI JIU 101 was towing the CJ50 jack-up drilling platform H1418. The data was captured at 25 frames per second with a resolution of 1920x1080 pixels. No new datasets were introduced; rather, existing real-world operational data was utilized for evaluation.
Key Insights and Ideas
- Leveraging the Sea-Sky Line for Roll: The ingenious use of the sea-sky line as a proxy for the vessel's rolling angle is a key insight. This avoids the need for dedicated inertial measurement units (IMUs) for roll estimation, simplifying the system's hardware requirements.
- Targeted Noise Filtering for Maritime Imagery: The development of specific filters to address sea wave texture and stern spray is critical. These are common and significant challenges in maritime computer vision, and their effective removal is essential for reliable towing line detection.
- Combining 2D Image Angles with 3D Geometry: The paper demonstrates a practical method for translating 2D image-based measurements (towing line angle in pixels, sea-sky line angle) into meaningful 3D world coordinates, accounting for the vessel's orientation.
- Robustness to Environmental Conditions: The experimental results suggest that the proposed method performs well even on complex sea surfaces with waves and spray, highlighting its practical applicability.
Key Results
The experiments successfully demonstrated the method's ability to estimate the towing line angle in real-time. The estimated 3D global towing line angles ranged from -1.00° to 7.58°, which were found to be consistent with GPS predictions. The rolling angle of the towing vessel was also estimated, showing fluctuations within a range of -2.53° to -0.62° degrees. The paper presents comparative experiments showing that the proposed method outperforms other saliency detection and edge detection techniques in accurately identifying the towing line.
Limitations and Future Work
The authors acknowledge that the height and distance of the tug and the towed platform from the camera can influence the towing line angle calculation. This is identified as a key area for future research. Further work could also explore the impact of other environmental factors and refine the geometrical projection model for even greater accuracy.
Summary updated