动态场景下融合深度信息的实时语义SLAM方法  

Real-time semantic SLAM method fusing depth information in dynamic scenes

作　　者：严瀚宇 孙博 马天力[1] 高嵩[1] YAN Hanyu;SUN Bo;MA Tianli;GAO Song(School of Electronic Information Engineering,Xi’an Technological University,Xi’an 710021,China)

机构地区：[1]西安工业大学电子信息工程学院,陕西西安710021

出　　处：《传感器与微系统》2025年第2期139-142,共4页Transducer and Microsystem Technologies

基　　金：陕西省重点研发计划资助项目(2022GY-242);咸阳市科技计划资助项目(2021ZDYF-SF-0020)。

摘　　要：在动态语义同步定位与建图(SLAM)系统中,由于语义分割计算时间长,造成系统的实时性较差,难以满足载体导航以及定位的需要。针对这一问题,本文提出了一种动态场景下融合深度信息的语义SLAM方法框架。首先,采用MASK基于区域的卷积神经网络(R-CNN)对关键帧进行语义分割;然后,利用贝叶斯滤波算法更新特征点的移动概率;最后,融合图像深度信息,通过对深度图像聚类与重投影误差约束辅助去除场景中存在的动态特征点。实验结果表明:相比于ORB-SLAM3算法,本文算法在TUM RGB-D数据集上的定位精度提高了94%,相比于DynaSLAM算法,运行速度提升了93%。In the dynamic semantic simultaneous localization and mapping(SLAM)system,due to the long calculation time of semantic segmentation,the real-time performance of the system is poor,which is difficult to meet the needs of carrier navigation and positioning.Aiming at this problem,a semantic SLAM method framework that fuse depth information in dynamic scenes is proposed.Firstly,the MASK region-based convolutional neural network(R-CNN)is used to semantically segment the key frames,and then the Bayesian filtering algorithm is used to update the movement probability of the feature points.Finally,the image depth information is fused,and the dynamic feature points in the scene are removed by the depth image clustering and reprojection error constraint.The experimental results show that compared with the ORB-SLAM3 algorithm,the positioning precision of the proposed algorithm on the TUM RGB-D dataset is improved by 94%,and the running speed is improved by 93%compared with the DynaSLAM algorithm.

关 键 词：动态环境 语义信息 关键帧检测 移动概率 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]