Deformation estimation of industrial objects from a single image
Loading...
Date
2024-09
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of New Brunswick
Abstract
Deformations introduced during the manufacturing process of plastic components degrade the accuracy of their 3D geometric information, hindering computer vision-based inspection. This phenomenon is prevalent among the primary plastic products where the objects are devoid of texture. This work proposes a solution for the deformation estimation of texture-less plastic objects using only a single RGB image. This solution encompasses a unique image dataset of five deformed parts, including both real-world and synthetic images, a novel method for generating mesh labels, sequential deformation, and a training model based on graph convolution. The sequential deformation method overcomes the prevalent chamfer distance algorithm in generating precise mesh labels. The model achieves a sub-millimeter accuracy on synthetic images and approximately 2.0 mm on real images, with an average testing time of 1.5 s on the Google Colab’s resources. The model’s high precision and speed make it suitable for real-world applications.