Computer vision methods for underwater pipeline segmentation

Abstract

Underwater pipeline inspection is usually conducted by Remotely Operated Vehicles (ROVs) equipped mainly with optical and acoustic sensors. During long inspections periods and low visibility conditions, traditional visual inspection becomes a tedious job and can lead to operator misinterpretations. Therefore, the automation of this process involves an improvement in the maintenance of the pipelines. This work presents an underwater pipeline segmentation system for rigid pipelines using a monocular camera. A color based edge detector was proposed, taking advantage of the pipeline geometry restrictions, besides tracking information. Segmented pipelines were transformed into a 2D top view representation. The system was evaluated with a dataset containing 7808 images, manually annotated, acquired during real inspection tasks. The system reached 96.5% of detection rate and 96.3% of segmentation accuracy.