A Sensor-Based Robotic Line Scan System With Adaptive ROI for Inspection of Defects Over Convex Free-Form Specular Surfaces

Abstract

In this paper, we present a novel sensor-based system to perform defect inspection tasks automatically over free-form specular surfaces. The inspection procedure is performed by a robotic manipulator equipped with a line scanner system. Taking the geometric and optical parameters into consideration, our algorithm computes a flexible scanning path. Based on a mesh model, the system segments the convex surface into areas with similar curvatures, then adaptively adjusts the line sensor's scanning range to ensure the complete coverage of the surface. We consider the scanning efficiency through constrained optimization. To align the computed scanning path with the specular object, a real-time workpiece localization algorithm in a coarse-to-fine manner is developed. Our control framework synchronizes the motion of the manipulator and the acquisition of the line scanner for robotic inspection. We also provide an image processing pipeline for automatic defect detection. To validate the proposed formal methodology, we report a detailed experimental study with the developed robotic inspection prototype.

A Multisensor Interface to Improve the Learning Experience in Arc Welding Training Tasks

“…To address the above mentioned issues, in this paper we propose an automated extended reality (XR) training assistant platform that provides the user with a multimodal experience. The proposed system consists of a high dynamic range (HDR) camera [16], [17] to monitor the welding spot in real-time, an RGB-D sensor [18], [19] to capture the 3D geometry of the scene, and a VR headset to safely visualize the process; Images from these sensors [20] are registered into a common frame. The system uses 3D vision to detect the welding region and seam, and to automatically generate the desired welding path [21].…”

Automatic Optical Inspection System for Wire Color Sequence Detection

“…If the developed inspection system is slower than the original manual process, it can seriously slow down the production line of the factory. Compared to training neural network models, traditional feature extraction methods [ 4 , 5 , 6 ] have relatively low hardware requirements and fast computational speed but are less robust to the environment.…”