When performing a direct vertical connection (DVC), the gooseneck may suffer severe damage due to the transfer of the load acting on the flexible pipe to the VCM (vertical Connection Module) when the bend restrictor is locked. On the other hand, the bend restrictor is necessary to guarantee that the minimum bend radius (MBR) of the flexible pipe is not violated during the installation.

Nowadays there is no way to assess this information in real time, except visually, thus the operation is very dependent on the pipe-laying engineer experience. In critical situations, to mitigate risks, a team of specialists onshore is requested to carefully evaluate the situation, assessing the radius of curvature, running DVC simulations using as input the geometric properties of the line and the VCM. Those geometric properties are estimated from images acquired by a remotely operated vehicle (ROV). This process is exhausting and time consuming, reducing the overall efficiency and increasing operational costs. Furthermore, it is error-prone because it does not consider the perspective effects of the image projection and the distortion caused by wide-angle lenses.

This paper describes a methodology to help the pipe-laying engineer carry out the DVC operations in a safe way. It is based on a computer vision system, SOIS, to estimate the curvature of flexible pipes during DVC operations, in order to increase the operational efficiency, through the use of stereo rig cameras and some markers along the pipe. It is faster and has a lesser margin of error than the simulated and manual assessment. This system relies on a stereo rig of lowlight cameras and on an interleaved pattern of black and white markers painted over the pipe. The 3D reconstruction process takes into account radial and perspective distortions of the images, resulting in a more accurate 3D geometry of the pipe. The system accomplishes its task through a sequence of three distinct phases: calibration, detection, and estimation of radius of curvature.

The camera calibration process, conceived for the underwater scenario, allows us to remove the radial lens distortion and to identify the camera intrinsic parameters, necessary to map image coordinates into world coordinates. The proposed detection algorithm is robust to non-uniform illumination and occlusion effects caused by fishes and particles floating around. From the detected points in both cameras of the stereo rig, it is possible to reconstruct the 3D geometry of the pipe. The estimation of the radius of curvature is obtained by fitting a 3D version of a catenary curve through the reconstructed geometry of the pipe.

Experiments demonstrated that the radius of curvature estimate is within a 3% margin of error at the 90% level of confidence. Field experiments showed that the pipe can be detected up to a distance of 15 meters. It is foreseen that the new methodology will significantly improve operational safety and shorten the average time of DVC operations, thus reducing its costs. Ongoing activities include improvements in the calibration phase to make it easier to be performed, and usability studies that are being conducted to improve the user experience with the system, especially for non-specialists in computer vision techniques.

This content is only available via PDF.
You can access this article if you purchase or spend a download.