单目视觉与IMU融合的室内定位方法研究  

Research on Indoor Positioning Method Based on Fusion of Monocular Vision and IMU

作　　者：同志学[1] 段东昱 张学锋 康智强[1] TONG Zhi-xue;DUAN Dong-yu;ZHANG Xue-feng;KANG Zhi-qiang(School of Mechanical and Electrical Engineering,Xi'an University of Architectural Science and Technology,Xi'an Shaanxi 710055,China)

机构地区：[1]西安建筑科技大学机电工程学院,陕西西安710055

出　　处：《计算机仿真》2024年第8期491-496,共6页Computer Simulation

基　　金：陕西省自然科学面上项目(2019JM-286);西安市创新能力强基计划(2021JH-05-0071)。

摘　　要：为提高室内环境下移动载体自主定位的定位精度,提出一种融合单目视觉和IMU的自主定位算法。基于视觉SLAM框架VINS-Mono,首先提取图像点特征并进行光流跟踪,在此基础上同时提取图像的线特征并计算描述子进行匹配,对场景结构提供额外的约束;然后在后端将点、线特征和IMU信息进行整合并通过滑窗模型进行优化,以提高后端位姿估计的精度;最后通过仿真和智能小车进行实验验证。结果表明上述定位算法相较于VINS-Mono的定位误差均方根平均降低了8%~13%,证明所提算法达到了提升定位精度的效果且满足定位实时性要求。In order to improve the positioning accuracy of autonomous positioning of mobile carriers in indoor environment,an autonomous positioning algorithm integrating monocular vision and IMU is proposed.Based on the visual SLAM framework VINS-Mono,firstly,the image point features were extracted and optical flow tracking is performed.On this basis,the line features of the image were simultaneously extracted and the descriptors were calculated for matching,which provides additional constraints on the scene structure;The point,line features and IMU information were integrated and optimized through the sliding window model to improve the accuracy of the back-end pose estimation;Finally,the simulation and smart car were used for experimental verification.The results show that the root mean square error of the positioning algorithm in this paper is reduced by 8% to 13% on average compared with VINS-Mono,which proves that the algorithm can improve the positioning accuracy and meet the real-time positioning requirements.

关 键 词：自主定位 视觉定位 视觉-惯性融合 定位精度 

分 类 号：TP202.7[自动化与计算机技术—检测技术与自动化装置]