Structure-from-motion based image unwrapping and stitching for small bore pipe inspections

Zhang, Dayi and Jackson, William and Dobie, Gordon and West, Graeme and MacLeod, Charles Norman (2022) Structure-from-motion based image unwrapping and stitching for small bore pipe inspections. Computers in Industry, 139. 103664. ISSN 0166-3615 (https://doi.org/10.1016/j.compind.2022.103664)

[thumbnail of Zhang-etal-CI-2022-Structure-from-motion-based-image-unwrapping-and-stitching]
Preview
 Text. Filename: Zhang_etal_CI_2022_Structure_from_motion_based_image_unwrapping_and_stitching.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (7MB)| Preview

Abstract

Visual inspection is one of the most ubiquitous forms of non-destructive testing, being widely used in routine pipe inspections. For small bore pipes (centimetre diameter), inspectors often have a restricted field of view limiting overall image and inspection quality. Stitching multiple unwrapped images is a common inspection technique to provide a full view inspection image by combining multiple video frames together. A key challenge of this method is knowing the camera pose of each frame. Consequently, mechanical centralisers are often utilised to ensure the camera is located centrally. For the inspection of small-bore pipes, such mechanical centralisers are often too large to fit. This paper presents a post-processing, Structure-from-Motion (SfM) based approach to unwrap and stitch inspection images, captured by a manually deployed commercial videoscope. It advances state-of-the-art approaches which rely on the projection of a laser pattern into the field of view, thus reducing the equipment size. The process consists of camera pose estimation, preliminary point cloud generation, secondary fitting, images unwrapping and stitching to form an undistorted view of the pipe interior. Two industrial focussed demonstrators verified the successful implementation for small-bore pipe inspections. Whereby the new approach does not rely on image features to create the surface texture and is less sensitive to the image quality, more areas can be retrieved from inspections. The reconstructed area was increased by up to 87% using the new approach versus the conventional 3D model.

CORE (COnnecting REpositories)