東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Linked to FindBook

Google Book

Amazon

博客來

Self-Supervised Semantic Learning of LiDAR Point Clouds for Large-Scale Scene Understanding.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Self-Supervised Semantic Learning of LiDAR Point Clouds for Large-Scale Scene Understanding./
Author:	Zhang, Haowei.
Published:	Ann Arbor : ProQuest Dissertations & Theses, : 2021,
Description:	96 p.
Notes:	Source: Masters Abstracts International, Volume: 83-06.
Contained By:	Masters Abstracts International83-06.
Subject:	Robotics. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28769595
ISBN:	9798496554848

Self-Supervised Semantic Learning of LiDAR Point Clouds for Large-Scale Scene Understanding.
Zhang, Haowei.

Self-Supervised Semantic Learning of LiDAR Point Clouds for Large-Scale Scene Understanding. - Ann Arbor : ProQuest Dissertations & Theses, 2021 - 96 p.

Source: Masters Abstracts International, Volume: 83-06.

Thesis (M.A.S.)--University of Toronto (Canada), 2021.

This item must not be sold to any third party vendors.

Semantic segmentation is a challenging task in the robotic vision community to classify various objects in a scene. While recent works employ supervised learning approaches using hand-annotated examples, these training samples are often costly to obtain.In this thesis, we present a self-supervised semantic learning method for large-scale scene understanding. Our offline method retrieves point cloud annotations by combining a mapping and localization solution with ray-tracing algorithms. Through multi-session navigation experiences in the same environment, our method labels points into four semantic classes: ground, non-movable, long-term movable, and short-term movable. These semantic labels allow us to train a semantic segmentation network without any hand annotations, which can then be used online.For qualitative and quantitative analysis, we demonstrate our method on a simulation dataset. We also provide a qualitative evaluation of a real-world dataset. Furthermore, by semantically filtering out movable points, results show that our method improves existing localization performance.

ISBN: 9798496554848Subjects--Topical Terms:

519753
Robotics.
Subjects--Index Terms:

Deep learning

Self-Supervised Semantic Learning of LiDAR Point Clouds for Large-Scale Scene Understanding.
LDR:02214nmm a2200361 4500 001 2343353
005 20220502104234.5
008 241004s2021 ||||||||||||||||| ||eng d
020 $a 9798496554848
035 $a (MiAaPQ)AAI28769595
035 $a AAI28769595
040 $a MiAaPQ $c MiAaPQ
100 1 $a Zhang, Haowei. $3 3681893
245 1 0 $a Self-Supervised Semantic Learning of LiDAR Point Clouds for Large-Scale Scene Understanding.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2021
300 $a 96 p.
500 $a Source: Masters Abstracts International, Volume: 83-06.
500 $a Advisor: Barfoot, Tim.
502 $a Thesis (M.A.S.)--University of Toronto (Canada), 2021.
506 $a This item must not be sold to any third party vendors.
520 $a Semantic segmentation is a challenging task in the robotic vision community to classify various objects in a scene. While recent works employ supervised learning approaches using hand-annotated examples, these training samples are often costly to obtain.In this thesis, we present a self-supervised semantic learning method for large-scale scene understanding. Our offline method retrieves point cloud annotations by combining a mapping and localization solution with ray-tracing algorithms. Through multi-session navigation experiences in the same environment, our method labels points into four semantic classes: ground, non-movable, long-term movable, and short-term movable. These semantic labels allow us to train a semantic segmentation network without any hand annotations, which can then be used online.For qualitative and quantitative analysis, we demonstrate our method on a simulation dataset. We also provide a qualitative evaluation of a real-world dataset. Furthermore, by semantically filtering out movable points, results show that our method improves existing localization performance.
590 $a School code: 0779.
650 4 $a Robotics. $3 519753
650 4 $a Computer science. $3 523869
650 4 $a Atmospheric sciences. $3 3168354
653 $a Deep learning
653 $a Point cloud
653 $a Robotics
653 $a Semantic learning
690 $a 0771
690 $a 0984
690 $a 0725
710 2 $a University of Toronto (Canada). $b Aerospace Science and Engineering. $3 2105951
773 0 $t Masters Abstracts International $g 83-06.
790 $a 0779
791 $a M.A.S.
792 $a 2021
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=28769595