一种基于亚像素匹配的高精度视觉定位方法  

A High-precision Visual Positioning Method Based on Sub-pixel Matching

作　　者：高嘉瑜 袁苏哲 李斌[1] GAO Jia-yu;YUAN Su-zhe;LI Bin(The 20 Research Institute of China Electronics Technology Corporation, Xi'an 710068,China)

机构地区：[1]中国电子科技集团公司第二十研究所,西安710068

出　　处：《导航定位与授时》2020年第3期54-61,共8页Navigation Positioning and Timing

基　　金：陕西省自然科学基础研究计划项目(2019JQ-936);陕西省科技厅2019重点研发计划“无人系统”(2019ZDLGY03-07-02);陕西省科技厅2019重点研发计划“北斗导航智能终端”(2019ZDLGY08-02)。

摘　　要：大多数应用到无人机景象匹配导航中的匹配算法都只能满足像素级精度,且当无人机利用景象匹配定位时图像之间往往存在较大的视觉差异,这种情况下图像匹配精度低且性能会急剧下降。所以选用具有仿射不变特性的特征点进行匹配,结合高斯亚像素拟合原理及特征描述符简化方法,并采用渐进采样一致性(PROSAC)算法剔除误匹配点,最终实现亚像素级的景象匹配定位。实验结果表明,采用此方法可以大大提高图像匹配的精度,提高到0.05像素以内。Most of the matching algorithms applied can only meet the need of UAV scene matching navigation with pixel-level accuracy,and when the UAV uses scene matching for positioning,often there are large visual differences between the images,leading to low image matching accuracy and sharp decline in performance.Therefore,we select feature points with affine invariant characteristics for matching,combine the Gaussian sub-pixel fitting principle and feature descriptor simplification method,and use the PROSAC algorithm to eliminate mismatched points to finally achieve sub-pixel level scene matching positioning.Experimental results show that this method can greatly improve the accuracy of image matching to within 0.05 pixels.

关 键 词：视觉导航 亚像素图像配准 PROSAC算法 仿射不变特征 简化描述符 

分 类 号：V279[航空宇航科学与技术—飞行器设计]