Abstract: The application of digital image measurement technology to conventional geotechnical triaxial tests has successfully solved many issues in traditional deformation measurements, significantly enhancing measurement accuracy and efficiency. However, the diversity and gradient of gray level strata in high-pixel images complicate gradient iteration calculations. Additionally, multiple potential corner candidate locations near corner points directly affect sub-pixel corner point localization accuracy. To address the localization bias problem of the sub-pixel corner detection algorithm, a sub-pixel corner detection method based on corner correlation matching identification is proposed for the digital image measurement system developed for full-surface deformation of geotechnical triaxial test specimens. The method initially extracts whole-pixel corner coordinates from the original image using the Harris detection algorithm, then constructs fixed-size grayscale sub-areas centered on these corner points. Subpixel-level displacements of the initial whole-pixel corner points after deformation are obtained through an iterative process of correlation matching between subregions. In the simulated specimen affine deformation test, the improved algorithm has good affine invariance on the basis of inheriting the robust identification of corner features. The average absolute error is reduced by 80.5%, substantially improving the measurement accuracy of full-surface deformation of the specimen in triaxial test.

Key words: corner-point correlation matching, digital image measurement, localization bias, grayscale subregions