Validation and performance of standalone KAZE for real-time strain measurement: CPU-only implementation and comparison with digital image correlation

View at publisher Summarise with VIRAT
Authors: Mahmoud Hussein, Daniel Mathieu, Gildas L’Hostis, Bernard Durand
Cited by: 0
References: 50

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract Digital image correlation (DIC) is the benchmark for non-contact tracking of a specimen during mechanical testing. Real-time operation demands high-powerful computers and complementary approaches to initialize the correlation process under large material deformations (speeded-up robust feature (SURF), scale invariant feature transform (SIFT), and others). We assess whether using only keypoints from feature detectors can reliably track displacement fields in quasi-static tensile tests, to run on a standard CPU to minimize cost. First, we examine the keypoints from SURF, SIFT, AKAZE, and KAZE detectors on synthetic speckle image sets involving (i) rigid-body displacements (0.01–2 px) and (ii) deformations mimicking uniaxial tension/compression tests (±0.1% to ±3.6%). We then present a custom outlier-rejection pipeline for mismatches. Second, we use real images acquired at 2 Hz from tests on an aluminum-alloy specimen. KAZE feature matching is then selected as a cost-effective alternative for real-time displacement/strain measurement with a conventional camera. The KAZE parameters were tuned using a reduced design of experiments, confirming satisfactory computation time. Performance relative to DIC is quantified on the real case by (i) evaluating information from the evolution of the number of keypoints when using a single fixed reference image versus systematically updating it, and (ii) comparing local displacements and average strain values in the central region of the specimens. The results show that KAZE feature matching provides a credible, computationally economical alternative for in-situ monitoring of quasi-static tests, even in the presence of nonlinear plastic deformations.

Fields of Study & Subjects

Optical measurement and interference techniques Advanced Image and Video Retrieval Techniques Fatigue and fracture mechanics

Publication details

Journal / source
Measurement Science and Technology
Volume
37
Issue
12
Pages
126002–126002
Publication date
27 March 2026
Language
EN