Temporal sequence Digital Image Correlation based on RNN-Conv3D

Digital Image Correlation (DIC) is a non-contact optical measurement method widely used in material mechanics and solid mechanics. Recent research has overcome some of the limitations of traditional DIC methods through convolutional neural networks. However, these methods seem to overlook the temporal continuity of displacement. Therefore, this paper aims to develop a deep learning-based solution that considers time continuity to optimize the performance of DIC. Firstly, this paper designs a convolutional neural network called RC3D-DICNet for measuring displacement fields corresponding to speckle image sequences. Secondly, based on Hermite elements, a method is proposed to generate speckle image sequences and real displacement field sequences. This approach is compared the latest deep learning methods through comparative experiments and result analysis conducted on self-built datasets. The results indicate that our proposed method achieves the highest average measurement accuracy. Furthermore, through star displacement field measurement experiments, it was found that our method performs better in spatial resolution.