Issue 31

R.D.S.G. Campilho et alii, Frattura ed Integrità Strutturale, 31 (2015) 1-12; DOI: 10.3221/IGF-ESIS.31.01 6 where f is the region of the image I with the same size as t centred in the position ( u , v ). Calculating  for all the pixels of I results in a matrix, where the maximum absolute value yields the location of the region in I that has the highest correlation with t and, thus, the most likely location of p i in the next image. This is done for every one of the eight points identified in the first image. After successfully identifying all the points of the second image, new templates are computed from the second image to search for the eight points in the third image, and so on until processing all the images. Computation of  n The value of t A CT in real world units (mm) is calculated as follows 1 2 CT A 3 4    p p t d p p (5) For all trials, a region of length d =45 mm was used (Fig. 3). The pixel size was on average 0.024 mm and, thus, the estimated maximum error of the image acquisition process is ±0.012 mm. Finally,  n can be defined as CT n A A    t t (6) where t A is the theoretical design value of 1 mm. Since t A can show small variations due to the fabrication process, an adjustment to  n is also applied to make  n =0 at the beginning of the test. Fig. 4 gives an example of the evolution of  n for a selected test specimen of configuration 2 (with h =4 mm). Shown in the graphic are the raw curve, the 6 th degree fitting curve and the corrected polynomial and final curve, adjusted to make  n (testing time=0)=0. This polynomial adjustment is required to smooth the raw data and remove experimental measurement scatter, but also to cancel any eventual misalignment between glued scales in both adherends.  n = -4.2482E-17 t 6 + 3.5246E-13 t 5 - 1.5875E-10 t 4 + 2.3802E-08 t 3 - 7.1272E-07 t 2 + 1.3125E-04 t - 1.2653E-04 R² = 9.9307E-01 0.00 0.04 0.08 0.12 0.16 0 50 100 150 200 250 300  n [mm] testing time, t [s] Raw curve Adjusted curve Polinomial (Raw curve) y ial ( r e) Figure 4 : Evolution of  n for one test specimen: raw curve obtained from the optical method, polynomial fitting curve and corrected polynomial curve. Computation of  o  o is calculated as the angle between lines l 1 and l 2 (Fig. 5) . These lines could be directly calculated from points ( p 5 , p 6 ) and ( p 7 , p 8 ) respectively. However, for increasing robustness to small fluctuations of the point detection process, an image processing algorithm was used to extract the midline of the edge of the ruler that contains the pair of points in hand. In particular, a Difference of Gaussian filters was applied for enhancing the edges of the ruler, resulting in an image where pixels belonging to edges have high intensity values, while the remaining ones have low intensity (Fig. 5).