3D Scene understanding from LiDAR point clouds

Muhammad Ibrahim

Research output: ThesisDoctoral Thesis

296 Downloads (Pure)

Abstract

Scene understanding commonly employs vision cameras. Nevertheless, 2D representations lack the depth information required for 3D world. LiDAR point clouds offer a rich representation of the outdoor environment for scene understanding. This dissertation presents a comprehensive study on 3D scene understanding from LiDAR point clouds. It aims to develop accurate 3D scene understanding algorithms using deep neural networks. Leveraging advancements in deep learning, the dissertation explores various methodologies based on 2D and 3D convolutions, and Transformers to process 3D point cloud data. This thesis also introduces three LiDAR based outdoor datasets for scene understanding, accompanied by an innovative annotation tool.
Original languageEnglish
QualificationDoctor of Philosophy
Awarding Institution
  • The University of Western Australia
Supervisors/Advisors
  • Mian, Ajmal, Supervisor
  • Wise, Michael, Supervisor
  • Akhtar, Naveed, Supervisor
Award date19 Dec 2023
DOIs
Publication statusUnpublished - 2023

Fingerprint

Dive into the research topics of '3D Scene understanding from LiDAR point clouds'. Together they form a unique fingerprint.

Cite this