Stereo digital image correlation using binocular super-resolution

The spatial resolution and measurement accuracy of digital image correlation (DIC) method are limited by the camera resolution. It is determined by the hardware cost. However, in the current stereo DIC measurement, only the gray level or its gradient of two images are used for integer-pixel matching and sub-pixel optimization. It subconsciously takes the two images from different views as independent individuals, and then correlates them. However, the structural information has not been utilized. This previously neglected information can provide a new approach to improve the accuracy of DIC. The realization of binocular super-resolution often needs to be based on a relatively small parallax. And DIC method can obtain image window pairing with small parallax by pre-matching. It means that binocular super-resolution and stereo-DIC can complement information for each other. In this paper, the DIC method is used to match the whole pixel of the image, and the binocular super-resolution method based on deep learning is used to process the image matching pairs. Based on the accumulation of previous experiments, rich datasets containing different experimental scenes and different speckle patterns were collated and used. Further, DIC method can establish a training datasets with minimal parallax through integer-pixel matching method, so it can achieve very excellent super-resolution results. Experimental results show that super-resolution images with higher signal-to-noise ratio can be obtained. Additionally, it can effectively provide more image details, which will be conducive to the calculation accuracy and resolution of DIC.