单目视觉移动机器人高精度鲁棒定位方法  

High-precision and robust positioning method for monocular vision mobile robot

作　　者：王晓华[1] 邓越 王雪玉 陈伟重 WANG Xiaohua;DENG Yue;WANG Xueyu;CHEN Weizhong(School of Electronics and Information,Xi’an Polytechnic University,Xi’an 710048,China)

机构地区：[1]西安工程大学电子信息学院,陕西西安710048

出　　处：《西安工程大学学报》2025年第2期19-27,共9页Journal of Xi’an Polytechnic University

基　　金：陕西省自然科学基金项目(2024JC-YBQN-0671);碑林区应用技术研发储备工程项目(GX2306)。

摘　　要：移动机器人在使用单目视觉传感器进行同步定位与建图(simultaneous localization and mapping,SLAM)时,复杂环境中存在大量的光线变化或环境纹理稀疏情况,这是导致其定位不准确的主要因素。因此,文中对ORB-SLAM3系统中前端与定位环节进行改进,提升单目视觉移动机器人在复杂环境中的定位精度与鲁棒性。首先,提出区域动态特征概率阈值调整算法对SuperPoint网络进行改进,替换原ORB算法进行图像特征提取,从而获取鲁棒性更强且分布更均匀的视觉特征点;其次,提出共视匹配策略和动态窗口匹配策略,优化了视觉前端的特征匹配与跟踪算法,提升在稀疏纹理场景下的视觉跟踪性能;最后,结合所提改进算法与多传感器信息融合技术,构建了完整的定位系统框架,并在该系统上进行了单目视觉地面移动机器人定位实验。实验结果表明:改进后的算法在EuRoc数据集上的绝对轨迹误差相比ORB-SLAM3降低了8.6%;真实环境中,机器人绝对轨迹误差相比改进前降低了33.59%。When mobile robots use monocular vision sensors for simultaneous localization and mapping(SLAM),they encounter numerous challenges in complex environments characterized by frequent lighting changes and sparse environmental textures,leading to inaccuracies in positioning.Therefore,this article focuses on improving the front-end and positioning components of the ORB-SLAM3 system to enhance the accuracy and robustness of monocular vision mobile robots.Firstly,we proposed a regional dynamic feature probability threshold adjustment algorithm to enhance the SuperPoint network,replacing the original ORB algorithm for image feature extraction.This step was aimed to acquire more robust and evenly distributed visual feature points.Secondly,we introduced a common view matching strategy and dynamic window matching strategy,optimizing the feature matching and tracking algorithm of the visual front-end.This optimization significantly improved visual tracking performance in scenes with sparse textures.Finally,by combining the proposed improved algorithm with multi-sensor information fusion technology,a complete positioning system framework was constructed.An experiment was conducted using this system.The results demonstrate that the improved algorithm reduces the absolute trajectory error on the EuRoc dataset by 8.6%compared to ORB-SLAM3.In a real-world environment,the error of the robot is reduced by 33.59%compared to that before the improvement.

关 键 词：移动机器人 同步定位与建图(SLAM) ORB-SLAM3 SuperPoint 单目视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]