Improving Public Transit Accessibility for Blind Riders by Crowdsourcing Bus Stop Landmark Locations with Google Street View
Kotaro Hara, Shiri Azenkot, Megan Campbell, Cynthia L. Bennett, Vicki Le, Sean Pannella, Robert Moore, Kelly Minckler, Rochelle H. Ng, Jon E. Froehlich · 2013 · Proceedings of the 15th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS) · doi:10.1145/2513383.2513448
Summary
This paper introduces Bus Stop CSI (Crowdsourcing Streetview Inspections), a scalable method for collecting bus stop location and landmark descriptions by combining online crowdsourcing with Google Street View (GSV). Blind and low-vision bus riders rely heavily on physical landmarks — shelters, benches, trash cans, newspaper bins, poles — to locate and verify bus stops, but this information is rarely available through digital tools or transit agencies. The paper reports on three studies: (1) formative interviews with 18 people with visual impairments across urban, suburban, and small-town settings to understand bus stop navigation challenges; (2) a comparative study validating GSV as a viable data source by comparing physical bus stop audits with virtual GSV audits across four neighborhoods in Washington DC and Seattle (179 bus stops, 42.2 linear km); and (3) an online study with 153 Amazon Mechanical Turk crowd workers who used the Bus Stop CSI tool to label landmarks at 150 bus stop locations. The tool features an interactive GSV interface with Explorer Mode for navigating streets and Labeling Mode for clicking on landmarks, with four-stage interactive tutorials for training workers.
Key findings
The formative interviews confirmed that half of participants had difficulty finding bus stop locations, with 53% relying on asking sighted pedestrians. Participants identified shelters and benches as the most helpful landmarks, followed by trash cans, newspaper bins, and non-visual cues like sounds and smells from nearby businesses. The comparative study showed high correlation between physical and GSV audit datasets (Krippendorff's Alpha: 0.944 physical, 0.930 GSV), with Spearman rank correlations all statistically significant (p < 0.001) — bus stop shelters had the highest correlation (ρ=0.88) and bus stop signs the lowest (ρ=0.61). In the crowdsourcing study, 153 turkers completed 3,534 labeling tasks producing 11,130 landmark labels. Individual turker accuracy was 82.5%, rising to 87.3% with a simple 7-turker majority vote quality control scheme. Shelters (88.8%), mailbox/newspaper bins (88.6%), and benches (83.3%) had the highest accuracy rates, while traffic signs/other poles (66.2%) were hardest to classify. The median labeling time was 44.7 seconds per panoramic image.
Relevance
This research demonstrates a practical, scalable approach to a real accessibility problem: the lack of detailed bus stop information that blind and low-vision transit riders need. Unlike previous approaches that required in-situ mobile crowdsourcing from transit riders themselves, this method allows anyone with internet access to contribute bus stop audits remotely using GSV. The 82.5-87.3% accuracy achieved by minimally trained crowd workers suggests this approach could be deployed at city or regional scale to populate transit agency databases with landmark information. For accessibility practitioners, this work illustrates how combining existing infrastructure (GSV imagery), crowd labor platforms, and well-designed labeling tools can address information gaps that disproportionately affect people with disabilities. Limitations include GSV image staleness (1-2 years old), occlusion problems, and the 29 bus stops missing from Google's Transit API. The vision of integrating this data into location-aware mobile apps that describe nearby landmarks as a blind pedestrian approaches a bus stop remains compelling.
Tags: visual impairment · blindness · crowdsourcing · public transit · navigation · Google Street View · bus stops · landmarks · wayfinding