PSI - Issue 17

Francisco Barros et al. / Procedia Structural Integrity 17 (2019) 986–991 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

988

3

2.2. Camera calibration: structure from motion and scale calibration

Two frames were chosen from each video sequence in such a way that they form two stereo pairs, one in the reference state and the other in the loaded state, suitable for DIC analysis, as can be seen in Fig. 2.

(a)

(b)

(c)

(d)

Fig. 2. Images used for DIC analysis: (a, b) reference; (c, d) loaded.

A total of 155 frames from the video of the reference state, including the images chosen for the DIC analysis, and the two selected images of the loaded state were used as inputs for a SfM com putation using Wu’s VisualSfM tool (Wu 2013), which finds features that appear in several of the captured frames using the SIFT method (Lowe 2004). The algorithm then computes the positions of detected features in three-dimensional space as well as the positions, orientations and intrinsic parameters (focal length and first-order radial distortion coefficient) of the camera at the moment when each image was acquired. In the images of the loaded state, the acrylic plate was blurred through the use of a Gaussian filter, since its position in these images is not the same as in the reference state. The camera’s