基于改进Lucas-Kanade的亚像素级零件图像配准*

针对工业应用中零件图像配准存在的光照变化和缺少纹理信息的难题,提出了改进Lucas-Kanade的亚像素级零件图像配准算法。首先根据光照变化和几何变换模型构建了模板与待配准图像间的非线性最小二乘函数；然后依据两幅图像的方向向量一致性和边缘特征为函数添加权重,以减少冗余像素点；最后应用Levenberg-Marquardt(LM)算法解算函数最优解,以实现精确图像配准。使用500幅待配准图像进行实验,结果表明该算法对缺少纹理的零件具备光照不变性,配准正确率高且达到亚像素级精度,能够满足工业应用的鲁棒性和精度要求。

Subpixel object-image registration using improved Lucas-Kanade

Due to the problem of image registration caused by illumination change and lack of texture in industrial applications, the subpixel object-image registration algorithm using improved Lucas-Kanade was proposed. Firstly, an nonlinear objective function about the template and image for registration was build using illumination and geometric transformation model. Secondly, weights were added to the objective function according to the consistency of direction vector of two images as well as edge features in order to reduce the redundant points. Finally, Levenberg-Marquardt algorithm was applied to solve the objective function. Using 500 images to test the proposed algorithm, experimental results indicate that the proposed algorithm was robust to illumination change with high accuracy rate, and had subpixel translation and rotating accuracy. The proposed algorithm can satisfy robustness and subpixel accuracy requirements under the industry conditions.