基于方差分量估计的视觉定位方法  

Aerial Re-fueling Visual Localization Stochastic Model Approach Based on Variance Component Estimation

作　　者：周泽波 孙诗媛 田学海 刘芮宏 ZHOU Zebo;SUN Shiyuan;TIAN Xuehai;LIU Ruihong(School of Aeronautics&Astronautics,University of Electronic Science and Technology of China,Chengdu 611731,China)

机构地区：[1]电子科技大学航空航天学院,四川成都611731

出　　处：《探测与控制学报》2024年第3期109-114,120,共7页Journal of Detection & Control

摘　　要：针对视觉导航中存在成像特征点的非均匀误差会影响位姿计算准确度的问题,提出一种基于方差-协方差分量估计的视觉定位方法。首先,从理论角度出发分析相机测量过程中随机误差存在形式及影响因素;其次,向双目相机视觉及视觉/惯性融合模型添加随机模型优化模块,构建非线性优化系统;最后,使用KITTI数据集、EuRoC数据集进行测试,对数据进行基于划分的聚类处理,验证得到该方法能够更加有效精准地估计飞行目标的位姿信息,对随机误差具有良好的优化效果。To address the issue of non-uniform errors in imaging feature points affecting the accuracy of position calculation in visual navigation,a visual localization method based on variance-covariance component estimation was propose.This method analyzed the existence form of random errors and the influencing factors in the camera measurement process from the theoretical point of view.To create an effective solution,a random model optimization module was integrated into the binocular camera vision and vision/inertial fusion model,resulting in the development of a robust nonlinear optimization system.The algorithm was validated using the KITTI and EuRoC datasets,with data processing based on division-based clustering.Experimental results showed substantial improvements in optimizing random errors making it a promising enhancement for visual navigation systems.

关 键 词：视觉定位 方差分量估计 非线性优化 随机误差 聚类模型 

分 类 号：V249[航空宇航科学与技术—飞行器设计] TP391[自动化与计算机技术—计算机应用技术]