Sidewalk AI Scanner: Turning residents' smartphone videos into city-wide sidewalk accessibility maps

The problem

For most people, a sidewalk is either usable or it is not. For a wheelchair user, a narrow stretch can mean having to turn back. For a parent with a pushchair, a broken surface can mean stepping into the road. For an older resident, uneven paving can mean a serious risk of falling.

Small design details shape whether people can move safely and independently through their communities. Yet in many cities, information about sidewalk accessibility is incomplete. Assessing sidewalks properly takes time. Trained surveyors visit streets in person. They measure widths, record kerb heights, note obstacles and assess surface conditions. This work is careful and valuable. It creates detailed and reliable records. But it is also labour-intensive and costly. Surveys happen infrequently, or primarily only after complaints.

Conditions can change long before official records are updated. For local authorities planning inclusive transport networks, this creates a blind spot. Without consistent, city-wide data, it is difficult to see where barriers are concentrated, which routes are genuinely accessible, or how conditions are changing over time.

Decisions about investment, maintenance and enforcement risk being shaped by partial information. In many places, the issue is not a lack of commitment to accessibility. Public servants want to design streets that work for everyone. The constraint is practical: how to gather reliable, up-to-date information at scale without relying entirely on specialist fieldwork.

The solution

Researchers at the Massachusetts Institute of Technology (MIT) Senseable City Lab set out to tackle this challenge. The lab focuses on how digital technologies can help cities understand everyday urban life. Here, the question was practical and direct: if governments cannot survey every sidewalk themselves, could residents help collect the data?

Almost everyone carries a smartphone. It includes a camera, GPS and basic motion sensors. The team asked whether this everyday device could become a simple, shared tool for consistently mapping sidewalks. From that question, Sidewalk AI Scanner was developed.

Sidewalk AI Scanner is organised into two main parts: a data collection tool and an accessibility dashboard.

1. The data collection tool

Residents use the web app on their smartphones to record a short video while walking along a sidewalk.

The app provides clear, step-by-step guidance. It explains how to hold the phone, how to frame the sidewalk and how to move at a steady pace. Visual examples and short tutorials help users consistently record footage.

This standardisation matters. It helps ensure that videos collected by different people, in different places, can still be analysed reliably. Once the video is recorded, it is uploaded through the app.

The system then extracts frames from the footage and applies computer vision, a form of artificial intelligence trained to recognise features linked to accessibility. It measures how wide each sidewalk is and checks the surface for damage, obstructions, and whether ramps are present. These observations are converted into georeferenced data. In other words, each sidewalk segment is mapped and linked to specific accessibility indicators.

2. The accessibility dashboard

The second part of the web app is a public-facing accessibility dashboard. It displays a catalogue of cities mapped using this approach. For each city, users can see an overall accessibility level, calculated from the average condition of scanned sidewalks. At street level, individual sidewalk segments are colour-coded on an interactive map. Green indicates accessible conditions. Red signals that one or more barriers have been detected.

The dashboard also shows which features were identified or are missing for each segment. Importantly, it indicates how recent the data is. Newer scans appear more clearly, while older ones fade. This helps users see where information may need updating.

Together, these two elements create a cycle: collect, analyse, visualise.

The aim is not to replace professional surveys. It is to widen the evidence base, giving cities a clearer, more up-to-date picture of sidewalk conditions across far more streets than traditional methods alone can cover.

The impact

1. A more complete, city-wide picture of accessibility

Traditional sidewalk surveys often cover only select streets, leaving large gaps in understanding. Sidewalk AI Scanner's participatory model enables data to be collected much more widely, helping to build a comprehensive view of sidewalk conditions. This makes it easier for planners to spot where barriers cluster and where people are most likely to face difficulties getting around.

2. A lower-cost, scalable alternative that still delivers useful insight

The approach combines crowdsourced imagery with automated analysis. It does not replace specialist survey equipment, but it gives cities an inexpensive way to gather structured citywide sidewalk data at scale. This means even councils with limited budgets can begin to fill long-standing data gaps over time.

3. A tool for community engagement and empowerment

Because the system uses smartphones, something many people already carry, it invites residents to actively map their own streets. This shifts sidewalk data collection from a specialist task to a shared civic effort, creating opportunities for deeper community involvement in transport planning. When residents see a tool that reflects their experiences, they can contribute directly to evidence that shapes decisions.

4. New perspectives on inclusivity and planning priorities

By generating consistent, mapped indicators of sidewalk features that matter for accessibility, the tool offers planners a new avenue for comparing conditions across places and over time. This can support more transparent prioritisation of investment and help track the impact of changes or upgrades. Rather than relying solely on ad hoc complaints or isolated audits, authorities can identify patterns from a broader evidence base.

Lessons learned

• Start with clear, shared definitions. Before building Sidewalk AI, the researchers reviewed existing accessibility standards and studies. They agreed on a defined list of sidewalk features that matter, such as width, slope, surface condition, obstacles and kerb ramps.

• Participatory data collection can be scaled with everyday devices. The team compared several data collection methods, including advanced 3D scanning. While these can produce highly detailed data, they are expensive and require specialist skills. Smartphone video, by contrast, is widely available and easy to use. Choosing this method made it possible for non-experts to contribute.

• Standardisation matters. To make crowdsourced footage useful, the system includes clear guidelines and best practices for recording. Without this structure, footage might be inconsistent and harder for the AI to interpret.

• Sidewalk data needs to reflect change over time. Sidewalk conditions are not fixed. Surfaces deteriorate. Obstacles appear and disappear. The dashboard includes indicators showing how each scan is dated.

Launch year: 2024

This case study was written with assistance from artificial intelligence.