DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM

Ryo Hachiuma, Christian Pirchheim, Dieter Schmalstieg, Hideo Saito

Research output: Chapter in Book/Report/Conference proceedingConference contributionResearchpeer-review

Abstract

We present DetectFusion, an RGB-D SLAM system that runs in real time and can robustly handle semantically known and unknown objects that can move dynamically in the scene. Our system detects, segments and assigns semantic class labels to known objects in the scene, while tracking and reconstructing them even when they move independently in front of the monocular camera. In contrast to related work, we achieve real-time computational performance on semantic instance segmentation with a novel method combining 2D object detection and 3D geometric segmentation. In addition, we propose a method for detecting and segmenting the motion of semantically unknown objects, thus further improving the accuracy of camera tracking and map reconstruction. We show that our method performs on par or better than previous work in terms of localization and object reconstruction accuracy, while achieving about 20 FPS even if the objects are segmented in each frame.
Original languageEnglish
Title of host publicationProceedings British Machine Vision Conference (BMVC)
Publication statusPublished - 2019
EventBritish Machine Vision Conference - Cardiff University, Cardiff , United Kingdom
Duration: 9 Sep 201912 Sep 2019
https://bmvc2019.org/

Conference

ConferenceBritish Machine Vision Conference
Abbreviated titleBMVC
CountryUnited Kingdom
CityCardiff
Period9/09/1912/09/19
Internet address

Fingerprint

Semantics
Cameras
Labels
Object detection

Cite this

Hachiuma, R., Pirchheim, C., Schmalstieg, D., & Saito, H. (2019). DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM. In Proceedings British Machine Vision Conference (BMVC)

DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM. / Hachiuma, Ryo; Pirchheim, Christian; Schmalstieg, Dieter; Saito, Hideo.

Proceedings British Machine Vision Conference (BMVC). 2019.

Research output: Chapter in Book/Report/Conference proceedingConference contributionResearchpeer-review

Hachiuma, R, Pirchheim, C, Schmalstieg, D & Saito, H 2019, DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM. in Proceedings British Machine Vision Conference (BMVC). British Machine Vision Conference, Cardiff , United Kingdom, 9/09/19.
Hachiuma R, Pirchheim C, Schmalstieg D, Saito H. DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM. In Proceedings British Machine Vision Conference (BMVC). 2019
@inproceedings{2dfa8bb25b704e39a1e024387c0fc14c,
title = "DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM",
abstract = "We present DetectFusion, an RGB-D SLAM system that runs in real time and can robustly handle semantically known and unknown objects that can move dynamically in the scene. Our system detects, segments and assigns semantic class labels to known objects in the scene, while tracking and reconstructing them even when they move independently in front of the monocular camera. In contrast to related work, we achieve real-time computational performance on semantic instance segmentation with a novel method combining 2D object detection and 3D geometric segmentation. In addition, we propose a method for detecting and segmenting the motion of semantically unknown objects, thus further improving the accuracy of camera tracking and map reconstruction. We show that our method performs on par or better than previous work in terms of localization and object reconstruction accuracy, while achieving about 20 FPS even if the objects are segmented in each frame.",
author = "Ryo Hachiuma and Christian Pirchheim and Dieter Schmalstieg and Hideo Saito",
year = "2019",
language = "English",
booktitle = "Proceedings British Machine Vision Conference (BMVC)",

}

TY - GEN

T1 - DetectFusion: Detecting and Segmenting Both Known and Unknown Dynamic Objects in Real-time SLAM

AU - Hachiuma, Ryo

AU - Pirchheim, Christian

AU - Schmalstieg, Dieter

AU - Saito, Hideo

PY - 2019

Y1 - 2019

N2 - We present DetectFusion, an RGB-D SLAM system that runs in real time and can robustly handle semantically known and unknown objects that can move dynamically in the scene. Our system detects, segments and assigns semantic class labels to known objects in the scene, while tracking and reconstructing them even when they move independently in front of the monocular camera. In contrast to related work, we achieve real-time computational performance on semantic instance segmentation with a novel method combining 2D object detection and 3D geometric segmentation. In addition, we propose a method for detecting and segmenting the motion of semantically unknown objects, thus further improving the accuracy of camera tracking and map reconstruction. We show that our method performs on par or better than previous work in terms of localization and object reconstruction accuracy, while achieving about 20 FPS even if the objects are segmented in each frame.

AB - We present DetectFusion, an RGB-D SLAM system that runs in real time and can robustly handle semantically known and unknown objects that can move dynamically in the scene. Our system detects, segments and assigns semantic class labels to known objects in the scene, while tracking and reconstructing them even when they move independently in front of the monocular camera. In contrast to related work, we achieve real-time computational performance on semantic instance segmentation with a novel method combining 2D object detection and 3D geometric segmentation. In addition, we propose a method for detecting and segmenting the motion of semantically unknown objects, thus further improving the accuracy of camera tracking and map reconstruction. We show that our method performs on par or better than previous work in terms of localization and object reconstruction accuracy, while achieving about 20 FPS even if the objects are segmented in each frame.

UR - https://arxiv.org/abs/1907.09127

UR - https://www.youtube.com/watch?v=Ys3FXEP3A_4

M3 - Conference contribution

BT - Proceedings British Machine Vision Conference (BMVC)

ER -