PSI - Issue 13

Francisco Barros et al. / Procedia Structural Integrity 13 (2018) 1993–1998 Author name / Structural Integrity Procedia 00 (2018) 000–000

1994

2

1. Introduction 1.1. Recalibration of a stereo rig in Digital Image Correlation

The accuracy of a 3D Digital Image Correlation (DIC) measurement is highly dependant on accurate stereo camera system calibration, and for the calibration to remain correct, it must ensured that the cameras do not move and that their internal parameters remain unchanged throughout the duration of the test. However, this is not always feasible, as it can be impractical to keep a stereo system in a fixed position and setup. For example, if DIC is used on a field measurement where the observed displacements occurs on a long-term time span, such as a structural monitoring application, the camera system should ideally be taken to the measurement site each time the images are acquired. This does not require an investment in a stereo vision system fully dedicated to the measurement for a long period of time, and avoids the need to keep the camera setup fixed and in good operational conditions at field measurement locations. In addition, calibration through moveable patterns of known geometry may be difficult to perform repeatedly if the surface being measured is large or hard to reach. In order for such a workflow to be possible, a reference system is needed that can not only perform stereo calibration, but also calibrate the system’s world coordinates to match its configurations from previous instances of the DIC test. The calibration of a stereo rig using only image point correspondences, i.e., point coordinates in several images that are known to correspond to the same physical feature, cannot, in general, be performed if the cameras’ intrinsic parameters are unknown [1, 2]. Therefore, this reference system relies on the presence of control points, either naturally occurring or set up for the purpose, whose world coordinates are known and fixed [2]. Malesa and Kujawinska [3] have developed a system in which DIC measurements of the same surface using different camera calibrations can be compared, by placing a two-dimensional pattern of known position and orientation which surrounds the area of interest, as the positions of the cameras can be determined relative to the two-dimensional pattern, which is fixed in space. This, however, requires a new calibration of the stereo camera system every time the configuration of the cameras is changed. 1.2. Overview of the developed work This paper describes a method which enables unremitting stereo DIC measurements even after the camera positions or intrinsic parameters have been changed. It relies on the detection of fixed points outside the region of interest (ROI) and the computation of their world coordinates with a known calibration, such that they become control points for the recalibration of the system. The method was tested under laboratory conditions, using auxiliary speckle patterns applied on surfaces of objects placed near the ROI, and using the SURF feature detection algorithm [4] to detect distinctive features in the speckle pattern and match them between different images. Feature tracking methods have been successfully applied to speckle patterns by Charrett et al. [5] for the detection of rigid motion in laser speckle pattern correlation. The nature of the measured speckle properties in the work by Charrett et al. is similar to the nature of the properties measured here: they are not field variables whose value depends on location and is measured in regular intervals, but parameters in a global geometric transformation which applies to any point correspondence within the pattern. Therefore, measuring values at exact predetermined locations is unnecessary. Each of the cameras in the stereo rig is calibrated separately. Features are matched between three images: images from both cameras captured using the reference calibration and one image from the camera being calibrated under the new setup. The detected points’ world coordinates are obtained from the images taken with the known calibration, and the image coordinates of those points in the image from the new camera configuration are used to determine the new calibration parameters and the relative position in relation to the reference state. Then, when performing DIC on the images with the new calibration, the resulting point cloud is obtained using the computed parameters and then translated and rotated according to the change in the position of the cameras, such that point clouds obtained with different camera positions are expressed in the same coordinate system.