Vision-based line detection for underwater inspection of breakwater construction using an ROV

Abstract

Ropes are often laid on the sea floor to guide remotely operated vehicles (ROVs) in the underwater inspection of breakwater construction. This paper proposes an algorithm to enhance the reliability of efforts to detect a yellow guide rope in ROV images, particularly in a turbid underwater environment. The algorithm comprises three processing stages: target enhancement, edge detection, and line detection. We also sought to optimize the three process parameters employed in the algorithm: the chrominance component of images for target enhancement, the Otsu method for hysteresis thresholding, and the fraction of sampled edge points for line detection. During target enhancement, images sent back from the ROV are converted to blue chromaticity (Cb) of the YCbCr color space to enhance the contrast between the guide rope and background. Edge detection is enhanced by using the Otsu two-thresholding method to adaptively determine the value for hysteresis thresholding for use in a Canny detector. Using the probabilistic Hough transform, we achieved a correctness exceeding 95% in line detection for rope images in turbid water even when using random sampling in which edge points accounted for only 40% of the total.