在微装配中, 采用变焦变倍视觉系统可以有效解决测量范围与精度的矛盾, 但同时也引入了动态标定和实时自动聚焦的新问题。为此, 对变焦变倍显微视觉系统的标定和自动聚焦技术展开研究。在标定方面, 首先通过变倍率法完成图像主点的标定。基于平面靶标定法, 采用单视图单应矩阵分解对固定倍率下相机内外参数进行线性标定, 再引入畸变模型, 并由量子行为粒子群优化算法对标定结果进行非线性优化, 优化之后的最大反投影误差约为0.13 pixel, 平均反投影误差约为0.1 pixel。此外, 通过高斯曲线拟合完成对任意工作状态下视觉系统放大倍数的校准。在自动聚焦方面, 针对传统灰度梯度函数只考虑固定梯度方向且易受噪声影响的问题, 采用八邻域最大梯度阈值的自动调焦算法, 通过梯度阈值提高算法的抗噪性。与其他几种灰度梯度调焦函数相比, 该算法的单峰性好, 抗噪性强。

In microassembly, zoom vision system was usually used to solve the contradiction between measuring scope and accuracy. However, new problems of dynamic calibration and real-time automatic focusing were also introduced as a side-effect. For this reason, the calibration and automatic focusing technology of zoom micro-vision system were described. For calibration, the principal point of image was determined by method of convertible magnification. Based on planar target, the linear calibration under fixed magnification was firstly performed by homography matrix decomposition of single view. Then the distortion model and the quantum-behaved particle swarm optimization (QPSO) were employed sequentially to do nonlinear optimization for the linear calibration result. After nonlinear optimization, the maximum re-projection error was 0.13 pixel and the average re-projection error was about 0.1 pixel. Furthermore, the calibration for magnification at arbitrary working condition was completed by Gaussian curve fitting. For real-time automatic focusing, method of maximum gradient threshold of eight-neighborhood and gradient threshold were used, for the traditional gray gradient function only considered the fixed gradient direction and was susceptible to noise. Compared with other several gray gradient focusing function, this method had good unimodality and noise immunity.