Metrological characterization of a low-cost portable stereo vision system for dynamic measurements

In recent years, camera and computer systems have been miniaturized considerably. The market now offers several low-cost products to assemble a small scale portable stereo vision system which can be completely autonomous (since it can acquire and process image). However, the metrological characteristics of such a portable stereo vision device should be assessed and compared with the state of the art of stereo vision system. In this work a portable stereo vision system composed by small board cameras and a single board computer is metrologically characterized and validated with respect to fixed non-portable systems. The analysis focused on the application of such a stereo rig to DIC and stereoscopic measurements performed with commercial software. For these tests two algorithms with different correlation techniques are evaluated in Matlab and in Halcon, moreover, a 3D DIC analysis is conducted on a H-shape specimen. To perform a metrological characterization, the points clouds obtained from the triangulations are confined to a region of the scene containing only a part of the specimen, after that they are fitted with a LOWESS fitting model. The uncertainty is assessed by computing the discrepancy between the points of the cloud and the ones of the fitting model. In some algorithms more than others the errors distribute in a strongly periodic manner with a period linked to integer values of disparity. Eventually the experimental analysis demonstrated that the accuracy of such a lightweight stereo system is similar to the accuracy of traditional not-portable ones.