Author ORCID Identifier

https://orcid.org/0000-0002-9951-2019

Document Type

Theses, Ph.D

Disciplines

Computer Sciences

Publication Details

Thesis Submitted for the degree of Doctor of Philosophy. Technological University Dublin, April 2024.

doi:10.21427/7cbh-n404

Abstract

Navigating urban environments is challenging for People with Visual Impairments (PVI). Many PVI opt to use navigation systems, such as Google Maps. However, current navigation systems often lack information crucial for PVI, such as the presence of traffic lights or roadworks. Knowing these features would enable PVI to understand their surroundings better and choose routes that suit their preferences. For instance, 79.6\% of PVI surveyed in this thesis prefer road junctions (termed intersections) controlled by traffic lights.

This research contributes to advancing navigation support for PVI in urban environments by focusing on annotating maps with useful environmental information. It proposes adapted solutions to address their specific needs through a four-stage approach. The initial stage reviews the existing research on outdoor navigation systems for PVI to evaluate its efficacy and limitations in addressing their needs. This review introduces a new taxonomy for the main phases and tasks required to support PVI during outdoor navigation. It also highlights current gaps in the literature, such as not providing PVI with relevant information, for example reading signage or detecting traffic lights.

The second stage conducts a questionnaire-based study with 49 PVI from Vision Ireland \footnote{Vision Ireland is Ireland's national agency for supporting PVI to live confidently and independently For more information, visit \url{https://vi.ie/}} to understand their needs and perspectives. Based on the questionnaire, this research proposes a general solution to address the main limitations of current outdoor navigation systems. My PhD work focuses on a specific part of this solution. This thesis considers specific environmental information, such as identifying traffic lights at intersections and details about street intersections, to help PVI select routes.

The identification of crucial intersection and traffic light data is the main goal of the third stage, which also aims to support the annotation of maps with environmental features that facilitate navigation planning. A satellite image dataset is constructed due to the lack of benchmark datasets providing intersection locations and degree labels. The thesis develops a deep learning framework to automatically detect the location and degree of road intersections from satellite images. Then, this information is combined with data about traffic light locations to tell PVI which intersections have traffic lights and which do not.

Finally, the fourth stage involves developing a routing algorithm customized for PVI. This algorithm called SafeRoute aims to help individuals select the safest path based on their preferences, considering walking distance and danger levels of route segments such as intersections or steps. Based on the experiments, the SafeRoute algorithm, built on the Dijkstra routing algorithm, reduces danger levels for PVI by almost 42\% compared to the most direct path available.

While this thesis addresses methods for enhancing navigation information for end users with visual impairments, its approaches are quite general. The techniques developed, including annotating maps with built environment details and extracting information from satellite images, apply to related domains, extending the impact of the research beyond its immediate scope.

DOI

https://doi.org/10.21427/7cbh-n404

Funder

Science Foundation Ireland

Creative Commons License

Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License
This work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 4.0 International License.

Download
Contact the Author


Share

COinS