室内机器人动态SLAM技术  

Dynamic SLAM technology of indoor robots

作　　者：韩慧妍 韩方正 韩燮[1,2,3] 况立群 曹亚明[1,2,3] HAN Hui-yan;HAN Fang-zheng;HAN Xie;KUANG Li-qun;CAO Ya-ming(School of Computer Science and Technology,North University of China,Taiyuan 030051,China;Shanxi Provincial Key Laboratory of Machine Vision and Virtual Reality,Taiyuan 030051,China;Shanxi Province Visual Information Processing and Intelligent Robot Engineering Research Center,Taiyuan 030051,China)

机构地区：[1]中北大学计算机科学与技术学院,山西太原030051 [2]机器视觉与虚拟现实山西省重点实验室,山西太原030051 [3]山西省视觉信息处理及智能机器人工程研究中心,山西太原030051

出　　处：《计算机工程与设计》2024年第2期530-539,共10页Computer Engineering and Design

基　　金：国家自然科学基金项目(62272426);国家自然科学青年基金项目(62106238);山西省自然科学基金项目(202303021211153);山西省研究生教育创新基金项目(2022Y632)。

摘　　要：随着计算机视觉、深度学习的发展,基于视觉的静态SLAM研究不断改进,然而大多数SLAM算法存在静态假定不成立和累计漂移。针对上述问题,设计动态BN-SLAM算法,利用深度图像的几何信息对原始掩模进行修正,修正掩膜用于移除运动物体及其影响。设计加权RANSAC方法,求解摄像机局部位姿。在TUM数据集上的实验结果表明,BN-SLAM的ATE、平移RPE和旋转RPE的平均RMSE值分别为95.46%、92.45%和90.88%,平均S.D.值分别为94.88%、94.76%和92.80%,跟踪轨迹点结果的平均率为98.80%。在真实环境的实验结果表明,BN-SLAM能够剔除运动的人造成的地图污染。With the development of computer vision and deep learning,vision-based static SLAM research has been improved,however,most SLAM algorithms suffer from untenable static assumptions and cumulative drift.To address these problems,dynamic BN-SLAM algorithms were designed to correct the original mask using the geometric information of the depth image,and the correction mask was used to remove moving objects and their effects.The weighted RANSAC method was designed to solve the local camera position.Experimental results on the TUM dataset show that the average RMSE values of ATE,translational RPE and rotational RPE of BN-SLAM are 95.46%,92.45%and 90.88%,respectively.The average S.D.values are 94.88%,94.76%and 92.80%,respectively.The average rate of tracking the trajectory point results is 98.80%.Results of experiments in real environments show that the BN-SLAM is able to eliminate map contamination caused by moving people.

关 键 词：计算机视觉 深度学习 静态假定 累计漂移 深度图像 原始掩模 修正掩膜 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]