动态场景下基于DeepLabv3+的语义视觉SLAM  

Semantic visual SLAM based on DeepLabv3+in dynamic scenes

作　　者：靳德利 仉新 张旭阳 朱文辉 左依林 JIN Dei;ZHANG Xin;ZHANG Xuyang;ZHU Wenhui;ZUO Yiin(School of Mechanical Engineering,Shenyang Ligong University,Shenyang 110158,China)

机构地区：[1]沈阳理工大学机械工程学院,辽宁沈阳110158

出　　处：《通信与信息技术》2024年第6期29-34,共6页Communication & Information Technology

基　　金：辽宁省教育厅面上青年人才项目(项目编号:LJKZ0258);2020年辽宁省科技厅博士科研启动基金计划项目(项目编号:2022-BS-187);沈阳理工大学高层次人才科研支持计划(项目编号:1010147001012)。

摘　　要：同时定位与地图构建(Simultaneous Localization and Mapping,SLAM)在人工智能领域有着不可替代的作用。传统的视觉SLAM算法在静态环境下具有较好的稳定性,但在动态场景下的鲁棒性和准确性较差,影响了其定位精度。为了解决这个问题,提出了一种结合ORB-SLAM3、语义分割线程和几何线程的语义视觉SLAM方法,利用DeepLabv 3+语义分割网络来分割潜在的先验动态对象,再通过几何线程使用多视图几何方法来检测动态对象的运动状态信息,剔除掉动态对象上的特征点,使用剩余的静态特征点求解相机位姿。最后,提出了一种新的蜣螂优化方案,通过最优路径找到所有动态特征点的集合,避免遍历所有特征点,减少动态目标检测时间,提高系统的实时性。通过在公开数据集上进行实验,结果表明,与同类算法相比,本文提出的方法有效提高了系统在高动态环境下的定位精度,提高了系统的实时性。Simultaneous Localization and Mapping(SLAM)plays an irreplaceable role in the field of artificial intelligence.The traditional visual SLAM algorithm is stable assuming a static environment,but has lower robustness and accuracy in dynamic scenes,which affects its localization accuracy.In order to solve this problem,a semantic visual SLAM method is proposed that combining ORB-SLAM3,semantic segmentation thread and geometric thread.Using DeepLabv3+semantic segmentation network to segment potential a priori dynamic objects.Then,the geometry thread uses a multi-view geometry method to detect the motion state information of the dy-namic object.The feature points on the dynamic object are eliminated and the remaining static feature points are used to solve the cam-era pose.Finally,a new Optimization scheme of dung beetle is proposed to find the group of all dynamic feature points through the opti-mal path,and avoids traversing all the feature points to reduce the dynamic object detection time and improve the real-time perfor-mance of the system.By conducting experiments on public data sets,the results show that the method proposed in this paper effectively improves the positioning accuracy of the system in a high-dynamic environment compared with similar algorithms,and the real-time performance of the system is improved.

关 键 词：DeepLabv3+ 语义 高动态环境 蜣螂策略 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]