Enhancing Collaborative SLAM with Semantic Information for Human-Readable Mapping

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

Sähkötekniikan korkeakoulu | Master's thesis

Date

2024-08-19

Department

Major/Subject

Autonomous Systems

Mcode

ELEC3055

Degree programme

Master's Programme in ICT Innovation

Language

en

Pages

57

Series

Abstract

This thesis discusses the enhancement of Collaborative Simultaneous Localization and Mapping (C-SLAM) system based on visual data with semantic information to produce human-readable maps. This work covers three steps of the semantic information integration: acquisition, aggregation and post-processing. By applying a state-of-the-art semantic segmentation neural network on each robot's video stream, the system is capable of obtaining the semantic information from multiple sources and classifying all the 3D map points with relevant semantic labels. These points are then used to extract a floorplan of the environment. The system was tested in a real office environment using three handheld devices. The experiment showed that the system can build a semantic point cloud and accurately extract wall outlines, room clusters, and scene objects. The findings suggest that semantic integration into C-SLAM systems can bridge the gap between human and robotic understanding of mapped environments.

Description

Supervisor

Cserép, Máté

Thesis advisor

Zhou, Quan

Keywords

collaborative SLAM, visual SLAM, semantic scene understanding, simultaneous localization and mapping, visual transformers

Other note

Citation