An improved 3D SIFT applied to estimate volume displacement field

摘要: DVC method has great potential in calculating 3D displacement field and strain field, but it relies heavily on the initial value of displacement field, especially for complex and large deformation cases. 3D SIFT algorithm is a very effective method to obtain feature point matching results. However, although the current 3D SIFT algorithm can extract a large number of feature points, the number of points that can be matched is not enough. This is because the orientation of the key points is not properly calculated. Therefore, we use the matrix singular value decomposition (SVD) method instead of the matrix eigenvalue decomposition method to deal with the image structure tensor of the key point and use Gaussian smoothing to deal with the amplitude of the gradient to obtain a more reasonable and accurate key point direction. It was verified by CT triaxial test of granite residual soil. The results show that the improved 3D SIFT method increases the matching rate of feature points by 110%~180%, and the matching results provide more accurate and dense initial displacement field estimation results for the digital volume correlation (DVC) method.