基于改进爬山算法的微小零件亚像素级定位

Subpixel localization of micro-parts based on improved hill-climbing algorithm

摘要:

为提高微小零件亚像素级定位效果，采用改进爬山算法。将目标区域向外扩展获得调焦窗口跟踪漂移图像，边界由像距和视角决定；然后优化爬山搜索算法得到最优收敛解，二维图像信息熵构成调焦图像评价函数；改进Zernike矩算法，通过高阶矩的模代替边缘参数，减少了计算量，卷积窗口矩阵构成Zernike矩的差值，提高定位精确度；最后给出了算法流程。实验结果显示，改进Zernike矩偶模板比奇模板边缘亚像素定位误差小，相比空间矩算法、多项式拟合算法、Zernike矩算法、区域生长算法和模板匹配算法，对规则形状定位误差均值分别减少了43.24%,21.62%, 32.43%,27.03%和56.76%；对不规则形状定位误差均值分别减少了39.02%,20.15%,26.83%,24.39%和51.22%。本文算法定位精确度较高。

Abstract:

Improved hill climbing algorithm is adopted in order to improve the subpixel positioning effect of micro-parts. Firstly, the focusing window is obtained by expanding the target area outward, which tracks the drift image, the boundary is determined by image distance and view angle. Secondly, the optimal convergence solution is obtained by using the optimized hill-climbing search algorithm, two-dimensional image information entropy is constituted by the focusing image evaluation function. Thirdly, modulus of higher moment of improved Zernike moment algorithm replace the edge parameters, which cut the computation amount. The convolution window matrix of difference of Zernike moment is constructed to improve the positioning accuracy. Finally, the flow of the algorithm is given. The experimental results show that the Zernike moment-even template has smaller subpixel positioning error than odd template edge method does, the improved hill-climbing algorithm is decreased by 43.24%,21.62%,32.43%,27.03% and 56.76% respectively compared with spatial moment algorithm, polynomial fitting algorithm, Zernike moment algorithm, region growing algorithm and template matching algorithm about the mean of positioning error of regular shape; the improved hill-climbing algorithm is decreased by 39.02%,21.95%, 26.83%,24.39% and 51.22% respectively compared with spatial moment algorithm, polynomial fitting algorithm, Zernike moment algorithm, region growing algorithm and template matching algorithm about the mean of positioning error of irregular shape. The positioning accuracy of the improved hill-climbing algorithm is higher.