By Michael Brenneis
Project Sidewalk, newly launched in Seattle, is crowdsourcing the evaluation of sidewalks and ramps with the intent to help DOTs locate and prioritize needed repairs and improvements, educate the public, and collect data to train AI. Poorly planned sidewalks and ramps, those in disrepair or with other impediments can dramatically reduce the mobility of people with disabilities and decrease walking accessibility.
After a brief tutorial, and using Google Street View, users systematically click around the city, identifying and evaluating curb ramps, sidewalk obstacles, and uneven sidewalk surfaces. During the Washington, D.C., pilot users placed labels on “more than 205,000 good and bad pieces of sidewalk over 18 months” as reported by Crosscut. The Seattle and Newberg, OR, versions of Project Sidewalk got underway in April.
The gathered data could eventually be incorporated into interactive routing software such as Access Map, which is aimed primarily at helping sidewalk users maximize their mobility. Project Sidewalk hopes to make its data available to city maintenance and planning agencies to improve their operations. They also intend to build a dataset robust enough to use to “train machine learning algorithms to automatically find accessibility issues” in street view images. Project Sidewalk may have the added benefit of educating citizens about the impediments faced by those with mobility issues and engaging citizens in the cause of improving the infrastructure of their cities.
Crowdsourced human intelligence tasks, such as sidewalk evaluation, can be vulnerable to malicious intent. People can make mistakes. The data can be compromised, or of poor quality. As with any form of data collection by humans, various types of bias can be introduced. To combat this, Project Sidewalk includes a validation component where users can examine other users’ work. The developers have also conducted field reconnaissance, finding that users’ evaluations are about 72 percent correct. An interesting follow up would be to see if this number could be improved by increasing the amount of training received by users, or by introducing other safeguards against misevaluation.
Transportation agencies can be slow to adopt or adapt these cutting-edge technologies for their own uses, continuing to rely largely on field observation and physical audits to assess the condition of infrastructure. Concerns over data quality, completeness, and accuracy seem to be paramount. But crowdsourced data could be used in combination with primary city data, such as maintenance or asset condition records, to prioritize areas in need of further physical inspection.
It’s an exciting time for crowdsourcing in the transportation field. Apps such as Ride Spot, Carbin, Strava, and Placemeter, among others, are collecting crowdsourced data from user observations or smartphone sensors that developers can leverage in imaginative ways. Analysts have access to many crowdsourced data sets (Open Street Map, for example) that are very useful for research purposes. From routing cyclists to reduce traffic stress, to routing cars for fuel efficiency, developers are incorporating crowdsourced data to conserve resources and improve the mobility and experience of bicyclists and sidewalk users of all abilities.
Michael Brenneis is an Associate Researcher at SSTI.