基于视觉的移动机器人相对位姿优化估计  

Vision-based Relative Pose Estimation for Mobile Robots

作　　者：黄双发 周乐来 李贻斌[1,2] HUANG Shuangfa;ZHOU Lelai;LI Yibin(School of Control Science and Engineering,Shandong University,Jinan 250061,China;Engineering Research Center of Unmanned System,Ministry of Education,Jinan 250061,China)

机构地区：[1]山东大学控制科学与工程学院,济南250061 [2]智能无人系统教育部工程研究中心,济南250061

出　　处：《无人系统技术》2023年第4期104-112,共9页Unmanned Systems Technology

基　　金：国家自然科学基金面上项目(61973191)。

摘　　要：针对多机器人协同定位问题,开展了基于视觉靶标的多移动机器人相对位姿求解技术研究,设计了一种基于单目视觉的多机器人相对位姿估算系统。首先,考虑多移动机器人的定位自主性,建立了四个单目相机组合的全方位、无死角的视觉定位感知监测系统。其次,针对视觉靶标观测存在的系统误差及丢帧等随机误差,搭建车辆的阿克曼转向数学模型,提出了基于扩展卡尔曼滤波融合车辆里程计信息的相对位姿估算算法。最后,通过对视觉观测信息与车辆里程计信息进行融合估算优化,动态场景下定位精度相对于初始方案有了较大提高,同时不存在丢帧等的随机误差,提高了系统的定位精度与定位鲁棒性,证明了所提算法的有效性。通过与初始方法进行对比分析,文中提出的方法在动态场景下定位精度提高了20%~25%,文中设计的基于扩展卡尔曼滤波的相对位姿估算方法在动态场景下定位精度与鲁棒性更优。For the multi-robot cooperative localization problem,a research on the relative positional solution technology of multiple mobile robots based on visual targets is carried out,and a multi-robot relative positional estimation system based on monocular vision is designed.Firstly,considering the positioning autonomy of multi-mobile robots,an all-round and dead-angle-free visual positioning monitoring system with four monocular camera combinations is established.Then,the Ackermann steering mathematical model of the vehicle is built to address the systematic errors and random errors such as frame loss in visual target observation,and a relative position estimation algorithm based on extended Kalman filtering fused with vehicle odometer information is proposed.At last,by optimizing the fusion of visual observation information and vehicle odometer information,the localization accuracy in dynamic scenes is improved by 25%compared with the initial scheme,and there is no random error such as frame loss,which improves the localization accuracy and robustness of the system.The effectiveness of the algorithm proposed in this paper is proved.By comparing and analyzing with the initial method,the method proposed in this paper improves the positioning accuracy by 20%~25%in dynamic scenes,and the relative position estimation method based on extended Kalman filter designed in this paper has better positioning accuracy and robustness in dynamic scenes.

关 键 词：移动机器人 单目视觉 阿克曼模型 扩展卡尔曼滤波 位姿估计 融合滤波 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]