【泡泡一分钟】DynaSLAM：基于动态目标检测和背景修复的视觉SLAM

会员服务 ·

【泡泡一分钟】DynaSLAM：基于动态目标检测和背景修复的视觉SLAM

2019 年 1 月 27 日 泡泡机器人SLAM

每天一分钟，带你读遍机器人顶级会议文章

标题：DynaSLAM: Tracking, Mapping and Inpainting in Dynamic Scenes

作者：Berta Bescos, Jose M. F ´ acil, Javier Civera and Jos ´ e Neira

来源：2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

编译：颜青松

审核：陈世浪

欢迎个人转发朋友圈；其他机构或自媒体如需转载，后台留言申请授权

摘要

通常而言，SLAM都会假设场景是固定不变的；然而对于实际环境而言，该假设过于理想，限制了目前流行的SLAM系统在服务机器人和自动驾驶等领域的应用。

图1 具有动态内容的RGB-D输入帧

在本文中提出了DynaSLAM，其在ORB-SLAM2的基础上添加了动态目标检测和背景修复技术。本文综合使用了基于多视几何和深度学习的方法来检测动态目标，并在此基础上得到一个场景的静态地图，然后在此基础上再对输入帧进行背景修复，填补动态目标遮挡的区域。

图2 经过背景修复后的输入帧

DynaSLAM具有单目、双目和RGB-D三种运行状态，因此本文也分别在三类数据集上进行了测试，实验中充分研究了效率对精度的影响。实验结果表明，在动态场景下本文提出的算法的精度超过了传统视觉SLAM的精度。值得一提的是，本文的算法还能提供场景的静态地图，更能符合实际应用中长时间使用对地图的需求。

图3 静态场景地图和位姿轨迹

Abstract

The assumption of scene rigidity is typical inSLAM algorithms. Such a strong assumption limits the useof most visual SLAM systems in populated real-world environments,which are the target of several relevant applications likeservice robotics or autonomous vehicles.

In this paper we present DynaSLAM, a visual SLAM systemthat, building on ORB-SLAM2, adds the capabilities of dynamicobject detection and background inpainting. DynaSLAMis robust in dynamic scenarios for monocular, stereo andRGB-D configurations. We are capable of detecting the movingobjects either by multi-view geometry, deep learning or both.Having a static map of the scene allows inpainting the framebackground that has been occluded by such dynamic objects.

We evaluate our system in public monocular, stereo andRGB-D datasets. We study the impact of several accuracy/speedtrade-offs to assess the limits of the proposed methodology. DynaSLAMoutperforms the accuracy of standard visual SLAMbaselines in highly dynamic scenarios. And it also estimatesa map of the static parts of the scene, which is a must forlong-term applications in real-world environments.

如果你对本文感兴趣，想要下载完整文章进行阅读，可以关注【泡泡机器人SLAM】公众号（paopaorobot_slam）。

欢迎来到泡泡论坛，这里有大牛为你解答关于SLAM的任何疑惑。

有想问的问题，或者想刷帖回答问题，泡泡论坛欢迎你！

泡泡网站：www.paopaorobot.org

泡泡论坛：http://paopaorobot.org/forums/