RGB-based line segment detection and processing for fabric corner recognition and robotic grasping

Abstract To address the difficulty of deploying complex grasping point detection methods in fabric unfolding and folding tasks for smart homes and automated production lines, this paper presents an RGB-based robotic grasping system for fabric corner recognition, targeting small- to medium-sized solid-color fabrics. For thin and deformable fabrics, ordinary depth cameras often provide unstable local depth measurements at lifted or folded corners, making precise corner height estimation difficult, while introducing depth sensing also increases hardware complexity and computational cost. Therefore, this work adopts an RGB-only perception setting and builds on a line segment detector (LSD)-enhanced line perception pipeline and a Line Segment Classification, Deletion, and Completion Tool (LS-CDCT) to recover task-relevant line structures and corner cues from 2D images. Based on the recovered line segment chains, a geometric fabric corner detection algorithm is developed for robust recognition of visible fabric corners. The detected corner points are further integrated into a single-arm robotic system with a corner-driven grasping strategy for automatic fabric corner grasping. Experiments were conducted on 160 fabric images containing 489 visible fabric corners under different postures, backgrounds, lighting conditions, image acquisition settings, and image sizes. The proposed method achieves a corner recognition rate of 92.84%. In 100 recognition-and-grasping trials, the overall corner-grasping success rate reaches 92.00%, and the grasping success rate conditioned on correct corner recognition reaches 100.00%. The results demonstrate the effectiveness of the proposed line perception and structural recovery framework for lightweight robotic fabric handling.