Robust real-time detection of an underwater pipeline

Currently, the methods of inspection of underwater structures employ remotely operated vehicles, guided from a support vessel by human operators. The risk of losing concentration calls for the development of an intelligent vision, guidance and control system to support the human activity. The paper presents a robust system for the detection and the real-time tracking of submarine pipelines. An active vision system is proposed to predict changes in the scene, and to direct computational resources to confirm expectations by adapting the processing mode dynamically. The system originates from an image-processing algorithm that was previously developed by the authors to recognise the pipeline in the image plane. The accuracy of this algorithm has been enhanced by exploiting the temporal context in the image sequence. The disturbances on acquired images caused by motion are partially removed by a Kalman filter. The filter proves advantageous in supporting the guidance and control of the ROV, and in making the image-processing module itself more robust. Sequences of underwater images, acquired at a constant sampling frequency from T.V. cameras, are used together with synchronised navigation data to demonstrate the effectiveness of the system.