基于引力场模型的图像匹配方法

Method for Image Matching Based on Gravitation Field Model

针对无人机视觉辅助导航的自主着陆技术研究,提出了引力场模型的图像匹配新方法,直接建立失配代价势函数的梯度场。只要进入正确匹配位置的收敛域,就可以按照最优估计理论找到最速下降路径,从而找到正确匹配位置。选择不同的引力场模型,可以得到不同的收敛域。低阶模型收敛域较大,但匹配精度较低;高阶模型匹配精度较高,但收敛域较小。当收敛域较大时,可以采用较大的搜索步长寻找收敛域,提高图像匹配搜索阶段的效率。零阶模型与采用欧氏距离作为基本测度的Hausdorff距离方法一致,但其引力场不便于计算;非零阶模型由质点的引力场线性叠加就可以得到质点系的引力场。仿真实验表明搜索阶段采用二阶模型效果最好,达到匹配图像的精度要求。

This paper introduces a new method for image matching based on gravitation field model. We can build gradient field of the mismatch cost function directly. As long as in convergence domain of the correct matching position, we find the steepest descent path according to the optimum estimation theory and then find the correct matching position. With the different gravitational field model, we get different convergence domain. The lower order model gets the larger convergence domain but the lower matching precision. The higher order model gets the higher matching precision but the smaller convergence domain. With the larger convergence domain, we can search more efficiently in image matching using the larger step size. The zero-order model is equivalent to the Hausdorff distance using Euclidean distance as the basic distance measure, but it is difficult to obtain the gravitational field. The nonzero order model gets the gravitational field of particle system easily using the linear superposition of single particle gravitational fields. The second order model is most suitable for searching, which has been evaluated via computer simulations. The form of the second order model is as same as the law of gravitv.