What Can an Algorithm Tell Us About How People Perceive Streets?

This map of perceived safety of New York City streets capes was developed using an algorithm by researchers at MIT. Click to use the interactive map.
This map of perceived safety of New York City streetscapes was created using an algorithm developed by researchers at MIT. Click to use the interactive map.

What makes people feel that a street is safe, and what do those perceptions tell us about different streets? A group of researchers at MIT have developed a formula designed to approximate people’s subjective reactions to the way streets look. They hope it will help chart shifts in the quality of city environments over time and prove useful to urban planners and architects seeking to better understand what makes streets appealing.

Based on survey responses from almost 8,000 people, the research team developed an algorithm they are using to rank the perceived safety of every streetscape image provided by Google Maps in New York, Boston, Chicago and Detroit. Using the algorithm enables the MIT researchers to rate many more streets than if they had relied on human surveys. They hope to eventually make maps like the one above available in every city in the Northeast and Midwest. They call the tool Streetscore.

The MIT team says their algorithm is a reliable mimic of how humans perceive visual cues in urban environments. Using a 1 to 10 scale, 84 percent of the time it can successfully predict whether real people will rate a street on the low end (less than 4.5) or the high end (more than 5.5). The factors incorporated by the algorithm are not public at this time.

The perception of “safety” that Streetscore approximates is defined vaguely, since the survey doesn’t explicitly distinguish between traffic violence and violent crime. But the researchers say they have found a correlation between homicide rates and Streetscore ratings in New York. Meanwhile, a look at the New York City map reveals that some streets with high rates of traffic injuries and fatalities, like Queens Boulevard, rate poorly, while others, like the leafier Ocean Parkway in Brooklyn, rate well. Industrial zones and streets under elevated highways stand out as some of the lowest-scoring streets in New York.

The researchers note in their report [PDF] that “suburban houses with manicured lawns and streets lined with trees” tend to score highly. However, the New York City map shows that several extremely dense urban streets, such as Manhattan avenues, also get very high scores.

Here’s how a random sample of streets ranked, on a scale of 1 to 10, 10 being the safest.

Image: Streetscore
Image: Streetscore

You can contribute to the project and help improve the tool’s accuracy by evaluating street images in an online survey.

6 thoughts on What Can an Algorithm Tell Us About How People Perceive Streets?

  1. Maybe it’s just me, but I didn’t see any correspondence with how I actually feel on any of the streets I looked at.

  2. This might be an accurate algorithm for extrapolating what streets someone would perceive as safe. But what really makes a street safe and what people perceive as safe, are often very different. People might perceive bike infrastructure as adding danger; they often perceive traffic lights as adding safety; in these two cases, their perception would be wrong.

    But this algorithm and mapping still is interesting and can be useful in some ways. A guess a next step would be to overlay a map of perceived safe streets against a map showing where accidents and injuries have happened.

    Anyway, the more tools the better for the safer streets movement. More data and more ways of analyzing it hopefully can be used to push politicians in the right direction.

  3. It doesn’t seem terribly accurate, but I applaud the effort and hope they continue to improve the system!

    Some areas rated 2.0 seemed far more safe/pleasant than some 9.0s.
    At least all of the freeways are in the <1.0 category, so it nailed that part!

  4. Too bad the pictures used to create the algorithm were taken from a car, early in the morning. This just tells us how scary is an empty neighborhood to a driver..
    Or maybe it explains why drivers crash in certain locations: they are scared and want to get away faster… The “bonfires of the vanities ” theory or the reverse of traffic calming .
    I hope someone add traffic crashes to the computation..

  5. I live in NYC, but this project doesn’t have my neighborhood. How Accurate can this be if they don’t have all of NYC.

  6. Interesting. However, my block in Park Slope has scores that range from 2.1 to 9.9. That’s an awfully broad range.

Leave a Reply

Your email address will not be published. Required fields are marked *


Counting Bikes and Cars Without a Clipboard

Liberate yourself from government transportation data that doesn’t tell you what you need to know! Break the chains of ignorance about how streets in your town are being used! Declare your independence from five-year-old data sets in PDF spreadsheets! Advocates have for too long been at the mercy of the limited data on travel patterns they […]
The red areas mark close calls between drivers and pedestrians. Image via City of Bellevue

Can Algorithms Design Safer Intersections?

Cities and tech firms are deploying new technology to gauge risks at dangerous intersections. These sensors, cameras, and machine-learning algorithms are promising, especially when it comes to measuring close calls that don't result in crashes - but cities are still figuring out how they can use this information. In the meantime, there's no reason to wait on designing safe streets.

Every Traffic Fatality in the U.S. — Mapped

What is the most dangerous street in your region? Which one most needs improvements to protect cyclists? Where do drunk drivers do the most damage? Thanks to a new tool from New York City data scientist Max Galka, you can get a pretty good sense with a few clicks. Galka recently completed a map of every traffic […]