Abstract
Digital Image Correlation (DIC) is a well-established non-contact optical metrology method. It employs digital image analysis to extract the full-field displacements and strains that occur in objects subjected to external stresses. Despite recent DIC progress, many problematic areas which greatly affect accuracy and that can seldomly be avoided, received very little attention. Problems posed by the presence of sharp displacement discontinuities, reflections, object borders or edges can be linked to the analysed object's properties and deformation. Other problematic areas, such as image noise, localized reflections or shadows are related more to the image acquisition process. This paper proposes a new subset-based pixel-level robust DIC method for in-plane displacement measurement which addresses all of these problems in a straightforward and unified approach, significantly improving DIC measurement accuracy compared to classic approaches. The proposed approach minimizes a robust energy functional which adaptively weighs pixel differences in the motion estimation process. The aim is to limit the negative influence of pixels that present erroneous or inconsistent motions by enforcing local motion consistency. The proposed method is compared to the classic Newton-Raphson DIC method in terms of displacement accuracy in three experiments. The first experiment is numerical and presents three combined problems: sharp displacement discontinuities, missing image information and image noise. The second experiment is a real experiment in which a plastic specimen is developing a lateral crack due to the application of uniaxial stress. The region around the crack presents both reflections that saturate the image intensity levels leading to missing image information, as well as sharp motion discontinuities due to the plastic film rupturing. The third experiment compares the proposed and classic DIC approaches with generic computer vision optical flow methods using images from the popular Middlebury optical flow evaluation dataset. Results in all experiments clearly show the proposed method's improved measurement accuracy with respect to the classic approach considering the challenging conditions. Furthermore, in image areas where the classic approach completely fails to recover motion due to severe image de-correlation, the proposed method provides reliable results.
Highlights
Digital Image Correlation, shortly known as DIC, is a class of image analysis techniques introduced in the early ’80s [1,2,3] that find spatial correspondences between different digital images. It can be classified as a non-contact optical metrology method that employs digital image analysis to extract full-field deformation measurements of objects that are subjected to external stresses
Quadratic similarity criteria are defined in literature as either sums of squared differences (SSD), cross-correlation criteria (CC) or variations which use normalized image intensities [18, 19].These require explicit modelling of the factors that might impact accuracy because they assign an equal importance to all pixels of a reference and deformed subset pair in the motion estimation process
This is an obvious limitation since it is perfectly reasonable to assume that most pixels present motions that can be properly modelled, some do not. These pixel motions lower the accuracy of the final motion estimate in direct proportion to both their numbers and magnitudes. This paper addresses these limitations by introducing a new subset-based DIC approach that uses robust estimation [21, 22] to penalize intra-subset pixel differences between reference and deformed subsets
Summary
Digital Image Correlation, shortly known as DIC, is a class of image analysis techniques introduced in the early ’80s [1,2,3] that find spatial correspondences between different digital images. Displacements are obtained by matching the image intensity pattern of the reference image to that of the deformed image This consists in the optimization of a similarity criterion that penalizes pixel differences between the two images, considering a pre-defined displacement model. Quadratic similarity criteria are defined in literature as either sums of squared differences (SSD), cross-correlation criteria (CC) or variations which use normalized image intensities [18, 19].These require explicit modelling of the factors that might impact accuracy because they assign an equal importance to all pixels of a reference and deformed subset pair in the motion estimation process.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
