Real-time vision-based tracking of submarine-cables for AUV/ROV

This paper describes a software that traces submarine cables in image sequences. At first, it creates a set of three kinds of filtered images from the original one. Next, by applying thresholds to them, an edge image is extracted. Thresholds are adjusted dynamically and adaptively. From results of Hough transform of the edge image, candidates of cable edge are selected. Then, likelihood of the candidates to be a line is evaluated. Finally, distance of the candidates to the cable in the previous image is evaluated. The software was successfully tested for its performance by applying it to the image sequences of actual submarine cables. It was also shown by experiments that, based on the cable position estimated by the software, the ROV (remotely operated vehicle) could trace a curved cable laid on the bottom of the test tank.