Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/222819
Title: USABILITY OF STREET VIEW IMAGERY IN ASSESSING BIKEABILITY
Authors: KOICHI ITO
Keywords: 2020-2021
Architecture
Master's
MASTER OF URBAN PLANNING
Filip Biljecki
Issue Date: 25-Jun-2021
Citation: KOICHI ITO (2021-06-25). USABILITY OF STREET VIEW IMAGERY IN ASSESSING BIKEABILITY. ScholarBank@NUS Repository.
Abstract: Bicycles are an essential part of urban transport in creating environmentally sustain- able and healthy cities. To quantitatively measure the level of cycling facilitation, many studies have developed methods to score bikeability in cities. As technologies advanced, traditional data collection methods, such as field surveys, have been sup- plemented by virtual auditing and crowd-sourced data collection, becoming more standardized, spatially granular, and less time- and resource-consuming. In recent years, many urban studies have utilized street-view imagery (SVI) and computer vision (CV) techniques to enjoy its benefits of accessibility to data and scalability of research. Although some studies have explored the usability of such innovations in data collection and analysis, those studies only focused on specific aspects of bike- ability. Applications of these technologies have not yet taken place in comprehensive bikeability assessment studies, possibly due to difficulties of using CV techniques for traditional urban transport researchers and the complexities of developing a comprehensive bikeability index system for CV researchers. Thus, this study aimed to fill the research gap by answering the following research questions. Can SVI and virtual audits replace traditional methods in assessing bikeability? If yes, then how much do SVI indicators contribute to the overall index? Can SVI indicators be used alone to assess bikeability? To examine the usability of SVI and CV techniques for bikeability assessment, this study selected Singapore and Tokyo as study areas and developed an index of bikeability, which are consists of 34 items under five categories based on SVI and non-SVI indicators used by previous studies, namely, connectivity, environment, infrastructure, perception, and traffic condition. Contributions of this study are not only the utilization of SVI indicators but also the proposal of using predictive modeling for perception scores by conducting online surveys for sample images on how people perceive streets by asking them to rate streetscapes in GSV images and building predictive models with high- and low-level features extracted from GSV images. Through exploratory analysis, it was found that, out of 3,633 unique pairs, 1,162 pairs of visual features and perception scores are correlated with p-values lower than 0.05. Results show that SVI techniques and SVI can be used to evaluate bikeability comprehensively and that 85 percent of the overall index’s variance can be explained by SVI indicators, while non-SVI indicators explained only about 40 percent; therefore, SVI-indicators are paramount in contributing to the overall index. Despite their large contributions to the overall bikeability index, it might still be beneficial to keep some non-SVI indicators, such as slope, because they are easy to obtain and difficult to replace with features extracted from SVI. This study faced the following challenges: quality and availability of data for both SVI and non-SVI indicators, limitations due to programming skills and computa- tional requirements, and difficulty to measure the reliability of the index developed in this study. The new method proposed in this study contributed to exploring the potential to overcome the issues faced by previous studies, such as the scalability of bikeability assessment, and future studies are expected to overcome the challenges identified in this study by adding more data sources, improving feature extraction method, and conducting a validation study.
URI: https://scholarbank.nus.edu.sg/handle/10635/222819
Appears in Collections:Master's Theses (Restricted)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
Koichi Ito 2020-2021.pdf14.53 MBAdobe PDF

RESTRICTED

NoneLog In

Page view(s)

36
checked on Jan 26, 2023

Download(s)

5
checked on Jan 26, 2023

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.