融合视觉显著性的无人机SLAM导航定位  

UAV visual SLAM navigation and positioning with visual salience

作　　者：黄龙杨[1] 王致远 屈若锟 熊乾凯 李诚龙 HUANG Long-yang;WANG Zhi-yuan;QU Ruo-kun;XIONG Qian-kai;LI Cheng-long(Air Traffic Management College,Civil Aviation Flight University of China,Guanghan 618307,China)

机构地区：[1]中国民用航空飞行学院空中交通管理学院,四川广汉618307

出　　处：《计算机工程与设计》2025年第2期562-569,共8页Computer Engineering and Design

基　　金：民航飞行技术与飞行安全重点实验室开放基金项目(FZ2021KF13);中央高校基本科研业务费专项资金基金项目(J2023-045);四川省大学生创新创业训练计划基金项目(S202310624250)。

摘　　要：为解决无人机在室外复杂环境中飞行利用单目视觉同时定位与地图构建算法进行导航定位时选取特征点质量不高、噪声干扰,存在位姿估计误差过大的问题,提出一种加入图像多尺度分解进行双向迭代的视觉显著性处理线程,引入特征点稀疏性约束提高选取特征点的质量来提高计算精度。通过仿真实验分析该算法的鲁棒性与实时性。将无人机室外飞行实验结果与其它算法进行比较,验证了该算法在室外复杂环境中大幅提高了无人机位姿估计的准确度。To solve the problems of low quality of selected feature points,noise interference,and excessive pose estimation error when UAVs fly in complex outdoor environments using monocular visual SLAM algorithms for navigation and positioning,the multiple images scale decomposition was added to perform a bidirectional iterative visual saliency processing thread,and feature point sparse constraints were introduced to improve the quality of selected feature points and improve the calculation accuracy.The robustness and real-time performance of the algorithm were analyzed through simulation experiments.Results of the UAV outdoor flight experiment are compared with that of other algorithms,verifying that the algorithm greatly improves the accuracy of UAV pose estimation in complex outdoor environments.

关 键 词：无人机视觉导航定位 同步定位与地图构建 视觉显著性 稀疏性约束 单目视觉 室外场景 噪声干扰 算法鲁棒性 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]