Super-sensitivity full-field displacement measurement method based on convolutional variational autoencoder

Computer vision technology is an important means of non-contact displacement measurement, but its displacement measurement sensitivity is often limited by the rounding error of the camera during image acquisition. This error causes the image to retain only integer-bit pixel intensity values, making it difficult to characterize tiny displacements below the theoretical sensitivity limit. To address this problem, we propose a super-sensitive full-field displacement measurement method based on Convolutional Variational Autoencoder (CVAE). This method constructs a CVAE network model, introduces the standard deviation matrix as prior information, and designs a weighted reconstruction loss function to achieve accurate reconstruction of the decimal pixel intensity in the raw image data. The reconstructed image is processed in combination with the optical flow method to achieve super-sensitive full-field displacement measurement. The effectiveness of this method in super-sensitive displacement measurement is verified by numerical simulation and laboratory experiments, and it shows good robustness under partial occlusion and illumination changes. This method initially provides a feasible solution for high-precision full-field displacement measurement in complex scenes.