Researchers use satellite photos and AI to map crosswalks

Comment on this story


As cities seek to encourage people to switch from driving to more environmentally friendly modes of transportation, modest sidewalks and crosswalks play an important role. After all, walking is the original zero-carbon locomotion.

But in some cases, city officials lack the money to invest in sidewalks, or haven’t studied their pedestrian infrastructure to the same extent as roads and streets. Gathering data about sidewalks and crosswalks has historically been a labor-intensive task — taking months in some cases — but transportation researchers are increasingly looking to satellite imagery and artificial intelligence to speed up the work.

Satish Ukkusuri, a professor of civil engineering at Purdue University, and Rajat Verma, a graduate researcher, recently developed a model to quickly identify crosswalks in a city. They spoke to the Washington Post about their research and how studying crosswalks could help city leaders create more complete pedestrian networks.

The Post: How well do cities understand their pedestrian infrastructure compared to highways?

Satish Ukkusuri: Typically, we overinvest in highways in terms of driving infrastructure for cars and trucks, which is definitely the fuel of our economy. But cities – especially if we want to make cities liveable and sustainable – need a very good sustainable transport infrastructure, and that comes primarily from pedestrians. We saw this in Covid. People really wanted to use more green space. People started cycling more. However, cities are limited in how much pedestrian infrastructure they can actually provide. So this project really aims to fill the gap by first identifying where we are in terms of our pedestrian infrastructure in cities.

A city with a very good pedestrian infrastructure, as seen in both Asian and European cities, also leads to better health outcomes. It also leads to more economic opportunities for people to participate in and a better social environment for people to participate in many activities. But US cities are not very convenient for pedestrians. You can see this in the increasing number of pedestrian accidents that are currently happening post Covid. While there are many reasons why these types of accidents happen, one of the main causes is the lack of connected pedestrian infrastructure that really complements what we have in the road networks.

READ :  Enterprise Artificial Intelligence Market Size 2022 Trending

The Post: What has been the traditional approach to collecting sidewalk and crosswalk data?

Rajat Verma: Traditionally the MPOs, the urban planning organisations, either use existing inventories or what we have seen more often is this concept of walking audits where individuals go to that target region and then check the quality of the existing infrastructure to find the points, that they find suitable for development. They then digitize all of this manually collected data and then set about identifying candidate segments or candidate locations in which to make their investments. As you can imagine, this is quite a tedious task. They did so in 59 European cities in 2021. They reported that even trained analysts took nearly 12 months to go through and code each function. Here we found the opportunity to use our methods and data.

Pete Buttigieg says the US can fix the most dangerous thing in transportation

The Post: You used a combination of satellite imagery and artificial intelligence. Walk me through what you have done and how it works.

Verma: The study is divided into two main parts. The first part is to identify crosswalks using satellite imagery. And the second part is, once we’ve identified the crosswalks, how do we connect them to more traditional analysis that analysts do, typically in planning agencies. The first part is the heavy work, the AI ​​related work: how we can use images, how we can apply different deep learning methods.

Deep learning is a buzzword that is on everyone’s lips. It actually turned out to be one of the very, very effective methods in computer vision. Deep learning has almost completely revolutionized the field of computer vision, as you can just pour huge amounts of data – in this case you can insert thousands or millions of images – and the model will automatically identify the important information just from the data itself. You don’t even have to specify that a zebra crossing has certain characteristics.

READ :  Cognet9 and Solve.Care Partner to Provide Artificial Intelligence for Healthcare

When we look at automatic image tagging on Facebook, or when you upload your image to Google Photos, it’s determined to be you, or it’s your dog, or a family member. This is called “object recognition” and has improved by leaps and bounds in recent years.

The Post: How did you use county data to test your approach?

Verma: If we’re trying to prove the effectiveness or efficiency of an algorithm, we also have to show that it’s effective on a real data set. In this case, we have assumed that everything provided by the Washington, DC government is correct. And based on that, we’ve shown that this is about 93 percent to 100 percent effective at counting crosswalks.

Ukkusuri: One of the challenges in fine-tuning object detection and image processing algorithms for these types of problems is scaling. If you show a few images or just focus on one lane, it’s much easier to identify crosswalks. But when you want to do this on an entire city scale, the problem quickly becomes very unsolvable. For the Department of Transportation in Washington, DC, they make investments at the network level, that is, for the entire city.

We need to address the size issue so we can first look at the network aspect of the problem and identify the gaps in the pedestrian infrastructure. Secondly, we must also think about the transferability of these solutions. We need to be able to develop it for one city and then we should be able to take it to any other city and we should be able to find solutions with minimal training. We also need to be careful to validate these methods with what’s really happening on the ground, which is why we need ground truth information, which in this case is provided by the DC Department of Transportation.

READ :  SAS touts integration with Singapore toolkit to help operationalise responsible AI adoption

The US push to reduce road deaths is aimed at strong regional differences

The Post: You talked about other methods that might take months to collect this data. How long do you need now?

Verma: It’s very interesting because after we trained the model, it took us about 10 to 15 minutes to recognize all the crosswalks in Los Angeles and DC together.

The Post: You’re talking about pedestrian network augmentation as an application. What is that?

Ukkusuri: When we talk about the network layer of connectivity, we want people to be able to go from point A to point B without really having to worry about being hit by a car. Or you see long stretches where there are wandering deserts. I mean, there’s really no infrastructure there. When cities make decisions like this, they don’t really pay attention. I mean, they get pressure from their local district president, for example, and then they just build this infrastructure. We really haven’t seen if this crosswalk will allow access to other parts of the city. So that global view, that network view of this pedestrian infrastructure, is something that’s really important to us.

Sometimes the shortest way is to use existing facilities and these facilities can be parks and green spaces that are available. Without necessarily having sidewalks or crosswalks everywhere, you can use the infrastructure you already have and make minor changes to make it more pedestrian-friendly.

There are many planning organizations that do not have the resources to invest in this type of pedestrian study. I think we can offer added value here.