对数极坐标变换在景象匹配制导中的应用

Scene matching using log-polar transform in missile guiding

针对成像制导中对实时性和准确性的要求,提出了一种基于对数极坐标变换的图像抗旋转和尺度变化的LMA算法。在图像匹配过程中用LMA算法对两幅图像透视变换矩阵的8个未知参数进行拟合。匹配过程中用多分辨率金字塔法对图像进行分解,采用由粗到精的搜索策略,减少计算量并避免了误匹配。通过对照片进行匹配测试,证明了这种方法的有效性和实用价值。

Aiming at reaching the real-time and precision requirements in missile scene matching guiding, we bring forward a rotate and scale resistant method LMA （Levenberg-Marquardt Algorithm） based on log-polar transformation. LMA is used to fit the eight unknown parameters of the transformation matrices of two perspective images. Multi-resolution pyramid is used to decompose the image and by searching from the coarsest level to the finest level,the computational complexity is reduced and getting trapped in local minima is avoided. The experimental results show that this approach is efficient and useful.