基于遗传算法的数字图像相关变形初值估计

为了克服传统的基于牛顿-拉夫森迭代的数字图像相关法受迭代初值影响较大等问题,提出了一种结合遗传算法的数字图像相关法。以待测数据点为中心,选取邻域内的若干估值点,通过基于遗传算法的数字图像相关法匹配出变形前后估值点对坐标;随机选取不共线的3组或以上估值点对代入仿射变换模型,依据仿射变换结果估计变形初值,并作为牛顿-拉夫森迭代初值;最后结合牛顿-拉夫森迭代法计算亚像素位移值。结果表明,该方法的匹配时间相对传统方法平均降低37.54%,相较于传统的数字图像相关法在搜索性能、匹配精度等方面更加可靠。该研究为数字图像相关法中迭代初值优化效果对匹配速度和精度的影响提供了参考。

Initial estimation of digital image correlated deformation based on genetic algorithms

Traditional digital image correlation method based on Newton-Raphson iteration is greatly influenced by the initial value of iteration. In order to overcome the problem, a digital image correlation method based on genetic algorithm was proposed. The data point to be measured was chosen as the center and several valuation points in the neighborhood were selected. The coordinates of estimated points before and after deformation were matched by digital image correlation method based on genetic algorithm. Three or more non-collinear pairs of valuation points were randomly selected to be substituted for the affine transformation model. The initial deformation value was estimated based on affine transformation results and it was used as the initial value of Newton-Raphson iteration. Finally, the sub-pixel displacement was calculated by Newton-Raphson iteration method. The results show that the matching time of this method is 37.54% less than that of the traditional method. Compared with the traditional digital image correlation method, it is more reliable in search performance and matching accuracy. This study provides a reference for the effect of iterative initial value optimization on matching speed and accuracy in digital image correlation method.