Robotic Perception of Pedestrians in Crowded Environments

Virgona, Alexander Joseph

Robotic Perception of Pedestrians in Crowded Environments

Virgona, Alexander Joseph

Permalink

Publication Type:: Thesis
Issue Date:: 2022

Open Access

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is open access.

Adobe PDF

Download contents and abstractAdobe PDF (242.59 kB)

Adobe PDF

Download thesisAdobe PDF (9.8 MB)

View statistics

Full metadata record

Field	Value	Language
dc.contributor.author	Virgona, Alexander Joseph
dc.date.accessioned	2024-02-20T02:38:57Z
dc.date.available	2024-02-20T02:38:57Z
dc.date.issued	2022
dc.identifier.uri	http://hdl.handle.net/10453/175758
dc.description	University of Technology Sydney. Faculty of Engineering and Information Technology.	en_US.UTF-8
dc.description.abstract	Robots have already increased the productivity of industries such as manufacturing, mining, and logistics by taking over many dangerous, dirty or dull tasks and freeing humans to focus on more interesting work. For the most part however, these robots are deployed in environments where they are isolated from humans; robots work best with other robots and machines. There remains an untapped potential for robotic technologies to enhance our daily lives and work collaboratively with us, but to do this safely and effectively they must be able to perceive humans in their environment. This is a challenging problem as humans can vary wildly in their appearance and as human environments are often dynamic and cluttered, a long way from the precisely controlled environment of the production line. The work presented in this thesis aims to enable robots and intelligent systems to better perceive humans through contributions to the core capabilities of detection and tracking. Considering that many human-robot interactions are likely to involve sharing walking space, this thesis considers these perception problems at the level of pedestrian interactions. A novel method for detecting the location and orientation of pedestrians from point-cloud data is presented which is able to handle occlusions of the lower body by virtue of focusing on the head and shoulders. Building on this detection capability, a tracking algorithm is proposed which leverages interpersonal distance constraints and assumptions about relationship between shoulder alignment and walking direction, to maintain robust estimates of the pose of all pedestrians in a crowded scene. The accuracy of the pedestrian pose detection algorithm is quantitatively evaluated by comparison with precise pose estimates from an optical motion tracking system. The outputs from the detection front-end are tracked using the proposed algorithm which is evaluated based on the CLEAR-MOT tracking metrics. Tracking performance is compared to a state-of-the-art tracking algorithm fed with the same detection inputs, showing improved performance under heavy crowding. Finally, a field study evaluates the tracking performance on real depth data captured in a busy inner city train station. The application of the technology has a patent, has been developed into a commercial product and is being trialled by a local government in Sydney, Australia as a congestion management tool. This showcases the applicability of this technology to enable the smart infrastructure of the future, able to perceive and therefore respond to human behaviour and better manage public space in our crowded cities.	en_US.UTF-8
dc.format	Thesis (ME)
dc.language.iso	en_US	en_US.UTF-8
dc.relation	https://opus.lib.uts.edu.au/bitstream/10453/175758/2/02whole.pdf
dc.rights	info:eu-repo/semantics/openAccess
dc.rights	The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
dc.rights	© 2022 Alexander Joseph Virgona
dc.rights	au.edu.uts.lib/cph
dc.title	Robotic Perception of Pedestrians in Crowded Environments	en_US.UTF-8
dc.type	Thesis
utslib.copyright.status	open_access	*

Abstract:

Robots have already increased the productivity of industries such as manufacturing, mining, and logistics by taking over many dangerous, dirty or dull tasks and freeing humans to focus on more interesting work. For the most part however, these robots are deployed in environments where they are isolated from humans; robots work best with other robots and machines. There remains an untapped potential for robotic technologies to enhance our daily lives and work collaboratively with us, but to do this safely and effectively they must be able to perceive humans in their environment. This is a challenging problem as humans can vary wildly in their appearance and as human environments are often dynamic and cluttered, a long way from the precisely controlled environment of the production line. The work presented in this thesis aims to enable robots and intelligent systems to better perceive humans through contributions to the core capabilities of detection and tracking. Considering that many human-robot interactions are likely to involve sharing walking space, this thesis considers these perception problems at the level of pedestrian interactions. A novel method for detecting the location and orientation of pedestrians from point-cloud data is presented which is able to handle occlusions of the lower body by virtue of focusing on the head and shoulders. Building on this detection capability, a tracking algorithm is proposed which leverages interpersonal distance constraints and assumptions about relationship between shoulder alignment and walking direction, to maintain robust estimates of the pose of all pedestrians in a crowded scene. The accuracy of the pedestrian pose detection algorithm is quantitatively evaluated by comparison with precise pose estimates from an optical motion tracking system. The outputs from the detection front-end are tracked using the proposed algorithm which is evaluated based on the CLEAR-MOT tracking metrics. Tracking performance is compared to a state-of-the-art tracking algorithm fed with the same detection inputs, showing improved performance under heavy crowding. Finally, a field study evaluates the tracking performance on real depth data captured in a busy inner city train station. The application of the technology has a patent, has been developed into a commercial product and is being trialled by a local government in Sydney, Australia as a congestion management tool. This showcases the applicability of this technology to enable the smart infrastructure of the future, able to perceive and therefore respond to human behaviour and better manage public space in our crowded cities.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/175758