An unsupervised digital image correlation method based on MaskFlownet

Digital image correlation (DIC) is an optical based method for measuring the full-field displacement field. In this study, an unsupervised digital image correlation is proposed by combining MaskFlownet and traditional displacement field estimation constraints. To reduce the training cost, a virtual dataset is rendered. In the dataset, a variety of speckle images and displacement fields are included to ensure the generalization and robustness of the network. Comparative experiment in the dataset shows that the measurement accuracy and calculation speed of the proposed method are better than the traditional method. Tensile experiment on metal sample proves that our method can be effectively used in practical condition.