基于鲁棒的模型约束构建大规模、无漂移的SLAM系统

每天一分钟，带你读遍机器人顶级会议文章

标题：Large-scale, Drift-free SLAM Using Highly Robustified Building Model Constraints

作者：Achkan Salehi, Vincent Gay-bellile, Steve Bourgeois, Nicolas Allezard, and Frederic Chausse

来源：IROS 2017 (International Conference on Intelligent Robots and Systems）

播音员：清蒸鱼

编译：鲁涛

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

今天给大家介绍一个基于模型约束实现的大规模、无漂移SLAM系统——Large-scale, Drift-free SLAM Using Highly Robustified Building Model Constraints，该工作发表在IROS2017。

为了实现大规模且有几何参照的SLAM系统，当前大多数单目（或其它形式的廉价传感器）解决方案的核心组件是基于关键帧的局部光束平差法（Bundle Adjustment）。这些方法一般只关注来自传感器数据之间的约束，却忽视了场景重建过程中产生结构约束（比如点云）的可能性。从方法论的角度来说，目前深度学习在SLAM领域的应用还很有限。作者考虑到上述两个问题，设计了一个快速深度学习网络，能同时推测环境的语义信息和结构信息，并基于贝叶斯框架将二者的结果传入光束平差过程中，从而为纹理匮乏的区域的3D点提供了更好的约束。

图1. 流程图。红色虚线框住的网络能够同时推测语义标签和面元的法向量。

图2.效果图。本文的SLAM系统建立了重建的点云和3D模型之间的约束，与同类方法相比，可以处理严重遮挡的情形。这是由于作者用深度学习同时获取了语义信息和结构信息，从而取得了十分可靠的数据关联。(a)(b)存在较多遮挡的例子；(c)部分实验结果的顶视图，其中包含两个闭环；(d)重建场景的3D示意图。

Abstract

Constrained key-frame based local bundle adjustment is at the core of many recent systems that address the problem of large-scale, georeferenced SLAM based on a monocular camera and on data from inexpensive sensors and/or databases. The majority of these methods, however, impose constraints that result from proprioceptive sensors (e.g. IMUs, GPS, Odometry) while ignoring the possibility of explicitly constraining the structure (e.g. point cloud) resulting from the reconstruction process. Moreover, research on on-line interactions between SLAM and deep learning methods remains scarce, and as a result, few SLAM systems take advantage of deep architectures. We explore both these areas in this work: we use a fast deep neural network to infer semantic and structural information about the environment, and using a

Bayesian framework, inject the results into a bundle adjustment process that constrains the 3d point cloud to texture-less 3d building models.

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。