SSF: segment skeleton filtering method to generate a clean direction for automatic blasting robots in shipbuilding

Abstract

A primary challenge in automating hull plate pretreatment is generating stable driving directions for blasting robots. This study proposes Segment Skeleton Filtering (SSF), a novel method to generate clean directions for blasting robots. The SSF method uses images collected by combining a camera and a laser-pattern device, consisting of three steps: semantic segmentation, skeletonization, and stable path generation (SPG). First, semantic segmentation employs a laser-pattern device to capture the curved profile features of welding beads and segments overlapping the region between the weld line and the unpretreated region using the early-branched U-Net. Second, skeletonization compresses the segmented region into a center region, reducing the segmentation noise and extracting features that are suitable for center line extraction. Finally, SPG extracts a straight line from this refined region using random sample consensus (RANSAC) and reduces angle variations via a Kalman filter to provide stable driving directions for blasting robots. An experimental validation was conducted using data obtained from the shipbuilding environment. The results demonstrate that the SSF method achieves superior performance concerning segmentation accuracy, robust centerline extraction, and stability in generating driving directions.