A key issue in experimental testing is to determine the three-dimensional shape of a test specimen and to follow and record how the shape of the specimen changes under imposed loading during the course of the experiment. In traditional testing the geometrical measurements are normally tracked by fixed strain gauges and inductive displacement transducers mounted on the surfaces of the test specimen, according to a pre-defined setup. Although these devices are extremely precise and reliable they suffer from being point-wise (strain gauge) or to have only one-dimensional measurement possibilities (distance between two points).

When large full-scale structures are set to test, e.g. in earthquake engineering, a rather complex and comprehensive set of measurement devices will be required in order to follow the two- or three-dimensional shape behaviour of the structures. These measurement systems are often costly and time-consuming to set up, and even comprehensive measurement systems will generally fall short when large numbers of measurement points distributed over large structural surfaces are required.

In recent years digital photogrammetry has emerged as an integral technique in structural testing due to its ability to simultaneously measure the deformation or displacements at an almost arbitrary number of locations over a certain structural specimen during a test campaign. This versatility of – point of interest – selection, the eventual densification of measurements and the fact that it is a non contact technique have given digital photogrammetry a prominent position in experimental testing at ELSA today.

The optical measurement techniques developed at ELSA aim to provide:

• Diagnostic on the boundary conditions of an experiment
• Optical measurements in substitution or in complement to the classical mechanical sensors
• Measurement fields of the displacements or of the strain
• Description of the cracking pattern of a specimen
• Segmentation of the architecture of the specimen

The digital photogrammetry/optical measurement techniques give also a posteriori information on the experiment, in zones where no sensors were installed. But these techniques need to be adapted to the working conditions of a civil engineering laboratory, characterised by a wide range of sizes and by variations of lighting conditions during an experiment. In addition, the applied techniques must be sufficiently accurate to detect small deformation on elements undergoing large displacements.

The quality of our cameras varies from low level, consumer cameras, which could be used when equipment may be destroyed, to medium video camera (cf. MEGAWIND (34MB)) and high dynamic cameras (cf. ESECMASE (28MB)).

Our measurements are made in mono and stereo-vision, using at most two cameras on the same scene. In this context, the calibration of the set-up is ideally accomplished using a calibration toolbox developed by Yves Bouguet from the Computer Vision Research Group, Department of Electrical Engineering, at the California Institute of Technology. A home-made adaptation to this toolbox permits the use all the reference points in the images, even if the calibrating object –a chessboard– obliquely intersects the frame.
The preparation of the specimen ranges from simple marking with targets to random texturing. Initially, networks of targets were merely stuck on zone of interest, and these targets were tracked by a periphery follower. The natural texture of the specimen could also be used to fit a family of geometrical transforms that could model the displacement as function of time (cf. MEGAWIND (34MB)). The present measurements use a random, dense artificial texture deposited on the specimen prior to the experiment.

The aim is to follow the same material points during the whole experiment, in order to have their displacement and deformation in a Lagrangian frame. Before each experiments, series of photos of the scene are taken in order to obtain, by averaging, a clean reference state which will be used as template for the tracking. In our actual tracking algorithm, this template is fitted by an interpolation function. In this way, the successive frames are compared to the analytic template, which can be strained and displaced in order to minimise the squared difference between current and reference state.

Some examples are exposed in what follows:

• MEGAWIND tests (34MB)
• ESECMASE tests (28MB).