Feature-based tracking with keypoint re-detection for local kinematic measurement in bending-under-tension tests: A comparative study with Lucas-Kanade optical flow
Gonzalo Cerezo-Calle, Javier Guilabert-Martinez, Luis M. Jimenez-Garcia, Miguel Angel Selles Canto, Carlos Perez-Vidal
Results in Engineering  (Diciembre 2026)
Ed. Elsevier  ISSN:2590-1230  DOI:https://doi.org/10.1016/j.rineng.2026.113298  - 32 (2026) 113298

Resumen:


Optical techniques are widely used in experimental mechanics for non-contact, full-field measurements of dis-placement and strain. While classical Digital Image Correlation and gradient-based optical flow methods such as Lucas–Kanade provide subpixel displacement estimates, feature-based tracking approaches based on keypoint descriptors (e.g., ORB, KAZE, MLDB, and SIFT) are often regarded as less suitable for strain-related measurements due to their discrete nature and limited displacement accuracy. However, they may offer improved robustness to point loss when combined with keypoint re-detection strategies.


 


This work presents a comparative evaluation of descriptor-based tracking pipelines with re-detection against a sparse Lucas–Kanade optical flow baseline for local kinematic measurement in bending-under-tension tests. The comparison focuses on temporal stability, spatial coherence of the estimated motion field, and the ability of each method to provide mechanically meaningful kinematic indicators near the pin–specimen contact.


 


The results show that Lucas–Kanade provides the smoothest and least noisy kinematic fields under favorable inter-frame conditions, confirming its value as a subpixel reference. KAZE and MLDB-based approaches emerge as competitive alternatives by offering a favourable balance between robustness and spatial coverage through re-detection. In contrast, ORB shows greater spatial heterogeneity, while SIFT is mainly limited by a lower density of valid correspondences. Overall, the study clarifies the trade-off between subpixel accuracy and robustness to point loss, supporting image-based tracking for local kinematic characterization in complex sheet-forming tests.