PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization

Wei Wang; Bing Wang; Peijun Zhao; Changhao Chen; Ronald Clark; Bo Yang; Andrew Markham; Niki Trigoni

doi:10.1109/JSEN.2021.3128683

PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization

Wei Wang, Bing Wang, Peijun Zhao, Changhao Chen, Ronald Clark, Bo Yang, Andrew Markham, Niki Trigoni

Research output: Journal article publication › Journal article › Academic research › peer-review

27 Citations (Scopus)

Abstract

In this paper, we present a novel end-to-end learning-based LiDAR sensor relocalization framework, termed PointLoc, which infers 6-DoF poses directly using only a single point cloud as input. Compared to visual sensor-based relocalization, LiDAR sensors can provide rich and robust geometric information about a scene. However, point clouds of LiDAR sensors are unordered and unstructured making it difficult to apply traditional deep learning regression models for this task. We address this issue by proposing a novel PointNet-style architecture with self-attention to efficiently estimate 6-DoF poses from 360° LiDAR sensor frames. Extensive experiments on recently released challenging Oxford Radar RobotCar dataset and real-world robot experiments demonstrate that the proposed method can achieve accurate relocalization performance.

Original language	English
Pages (from-to)	959-968
Number of pages	10
Journal	IEEE Sensors Journal
Volume	22
Issue number	1
DOIs	https://doi.org/10.1109/JSEN.2021.3128683
Publication status	Published - 1 Jan 2022
Externally published	Yes

Keywords

LiDAR point cloud
LiDAR sensor relocalization
sensor applications

ASJC Scopus subject areas

Instrumentation
Electrical and Electronic Engineering

More information

10.1109/JSEN.2021.3128683

Cite this

@article{19a176e281714320bf2b2a38fc535e7f,

title = "PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization",

abstract = "In this paper, we present a novel end-to-end learning-based LiDAR sensor relocalization framework, termed PointLoc, which infers 6-DoF poses directly using only a single point cloud as input. Compared to visual sensor-based relocalization, LiDAR sensors can provide rich and robust geometric information about a scene. However, point clouds of LiDAR sensors are unordered and unstructured making it difficult to apply traditional deep learning regression models for this task. We address this issue by proposing a novel PointNet-style architecture with self-attention to efficiently estimate 6-DoF poses from 360° LiDAR sensor frames. Extensive experiments on recently released challenging Oxford Radar RobotCar dataset and real-world robot experiments demonstrate that the proposed method can achieve accurate relocalization performance.",

keywords = "LiDAR point cloud, LiDAR sensor relocalization, sensor applications",

author = "Wei Wang and Bing Wang and Peijun Zhao and Changhao Chen and Ronald Clark and Bo Yang and Andrew Markham and Niki Trigoni",

note = "Publisher Copyright: {\textcopyright} 2001-2012 IEEE.",

year = "2022",

month = jan,

day = "1",

doi = "10.1109/JSEN.2021.3128683",

language = "English",

volume = "22",

pages = "959--968",

journal = "IEEE Sensors Journal",

issn = "1530-437X",

publisher = "IEEE",

number = "1",

}

TY - JOUR

T1 - PointLoc

T2 - Deep Pose Regressor for LiDAR Point Cloud Localization

AU - Wang, Wei

AU - Wang, Bing

AU - Zhao, Peijun

AU - Chen, Changhao

AU - Clark, Ronald

AU - Yang, Bo

AU - Markham, Andrew

AU - Trigoni, Niki

PY - 2022/1/1

Y1 - 2022/1/1

N2 - In this paper, we present a novel end-to-end learning-based LiDAR sensor relocalization framework, termed PointLoc, which infers 6-DoF poses directly using only a single point cloud as input. Compared to visual sensor-based relocalization, LiDAR sensors can provide rich and robust geometric information about a scene. However, point clouds of LiDAR sensors are unordered and unstructured making it difficult to apply traditional deep learning regression models for this task. We address this issue by proposing a novel PointNet-style architecture with self-attention to efficiently estimate 6-DoF poses from 360° LiDAR sensor frames. Extensive experiments on recently released challenging Oxford Radar RobotCar dataset and real-world robot experiments demonstrate that the proposed method can achieve accurate relocalization performance.

AB - In this paper, we present a novel end-to-end learning-based LiDAR sensor relocalization framework, termed PointLoc, which infers 6-DoF poses directly using only a single point cloud as input. Compared to visual sensor-based relocalization, LiDAR sensors can provide rich and robust geometric information about a scene. However, point clouds of LiDAR sensors are unordered and unstructured making it difficult to apply traditional deep learning regression models for this task. We address this issue by proposing a novel PointNet-style architecture with self-attention to efficiently estimate 6-DoF poses from 360° LiDAR sensor frames. Extensive experiments on recently released challenging Oxford Radar RobotCar dataset and real-world robot experiments demonstrate that the proposed method can achieve accurate relocalization performance.

KW - LiDAR point cloud

KW - LiDAR sensor relocalization

KW - sensor applications

UR - http://www.scopus.com/inward/record.url?scp=85122368511&partnerID=8YFLogxK

U2 - 10.1109/JSEN.2021.3128683

DO - 10.1109/JSEN.2021.3128683

M3 - Journal article

AN - SCOPUS:85122368511

SN - 1530-437X

VL - 22

SP - 959

EP - 968

JO - IEEE Sensors Journal

JF - IEEE Sensors Journal

IS - 1

ER -

PointLoc: Deep Pose Regressor for LiDAR Point Cloud Localization

Abstract

Keywords

ASJC Scopus subject areas

More information

Other files and links

Fingerprint

Cite this