International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD2023)

89 / 2023-09-17 18:33:28

Deformation initial value estimation method based on an improved SIFT algorithm

Digital Image Correlation,Initial value estimation,SIFT algorithm,Sub-pixel registration

Special Sessions > SS05 - Computer Vision for Automation Science and Engineering

In the digital image correlation  (DIC)  method, obtaining accurate initial deformation parameters is an important prerequisite for sub-pixel registration. This work proposes a deformation initial value estimation method based on an improved SIFT algorithm. The TBWD-SIFT feature descriptor is constructed by optimizing the SIFT feature descriptor. The sub-pixel registration is performed by the inverse combination Gauss-Newton method to obtain accurate deformation results. The experimental results show that the initial estimation based on the improved SIFT algorithm improves the quality of feature points. Compared with the traditional SIFT algorithm, the correct matching rate is increased by 1.7 % on average, and the matching time is reduced by 17.4 %. The pixel error of the digital image correlation method is reduced to 42.8 % of the traditional SIFT algorithm, and the system error is reduced by about one order of magnitude. This study can provide accurate and reliable initial deformation values for the sub-pixel registration algorithm. While accelerating the convergence speed of the sub-pixel registration algorithm, the accuracy of the deformation results is improved.