Artificial Intelligence is going to bring a sea change to society. While AI has a wide range of applications from detecting cancer to simplifying conversation between different language speakers, it could also have a big effect on measuring accessibility. While every country, state and city are understanding the impact of AI, the City of Amsterdam is going a step further to measure accessibility using AI.
Every day, people with disabilities encounter a number of challenges or barriers while moving around or participating in the city. With AI, the City of Amsterdam wants to find these barriers and solve the bottlenecks. It has partnered with World Enabled to build a process resulting in a “meaningful, useful and usable application of AI”.
How the City of Amsterdam will use AI to measure accessibility
AI can be classified into five major types: machine learning, computer vision, natural language processing, speech recognition and robotics. In order to measure accessibility and eliminate the bottlenecks, the City of Amsterdam is essentially relying on machine learning and computer vision.
For machine learning, the City of Amsterdam is building rich datasets using existing data and for computer vision, it is relying on both existing data and crowdsourced data. The fact that it could identify the meaningful data available at its disposal and its ability to build on top of it is commendable. From tackling the accessibility topic to creating a research methodology, we look at ways the City of Amsterdam is going ahead with its mission to measure accessibility using Artificial Intelligence.
AI to tackle the topics impacting accessibility
A recent study suggests that 25 per cent of all people encounter barriers in cities based on age or disability. Dr. Victor Pineda, President of World Enabled argues that we can “accelerate the development and deployment of new urban tech and G2B solutions at scale with right partnerships and tools”. It all starts with identifying a whole range of possibilities in tackling the impact of accessibility.
To kick start this development, the City of Amsterdam and World Enabled have gathered an international team of accessibility experts to understand which accessibility-related information could be extracted from street panorama images. This is an excellent example of solving a problem using existing data. This data is collected every year by the City of Amsterdam and saved after the removal of privacy-sensitive information such as license plates and people.
The experts were then divided into three groups and they were asked to do the following things:
- Annotate the selected images with a red sticky note for bad accessibility practices. A green sticky note was used for good practices and experts were also asked to add their explanation.
- Identify the type of accessibility-related information missing on the city panorama images
The exercise concluded that the City of Amsterdam could use the panorama images to identify the existence or lack of pedestrian features such as “obstacles, curb ramps, crossings, push buttons, tactile markings, street furniture, among others”. However, these images were not helpful to make “precise measurements, time predictions, non-visual information, buildings accessibility”.
The experts also deemed them less useful for identifying the contrast of pavements and signs could not be evaluated due to visual properties or the dynamic lighting of the images. The workshop allowed the City of Amsterdam to understand the broad spectrum of measuring accessibility for pedestrian mobility. To further the research, here is a look at the methodology followed and data sources collected:
- Collected the accessibility guidelines for designing a barrier-free public space from the municipality of Amsterdam and United Nations Enabled.
- Combined the results of the workshop and the accessibility guidelines. The results were then categorised and described for each accessibility-related feature.
- Creation of a survey for accessibility experts: In order to measure the level of impact for each feature, a survey was created looking at features that will answer the need of most disability types and/or the pain points most reported.
- Creation of a survey for data scientists: This allows to measure the feasibility of available datasets.
- Prioritising based on the survey’s results: The City of Amsterdam and World Enabled then selected the city features evaluated with the highest impact and feasibility.
Sidewalk and crosswalk accessibility assessment
It is important to understand that an AI program will be ineffective without sufficient or meaningful datasets. In its quest to measure accessibility, the City of Amsterdam and World Enabled managed to understand the pain point using datasets and then continued with its research on sidewalk and crosswalk accessibility assessment.
They learnt that people with disabilities frequently had trouble with obstacles on the sidewalk. They also learnt that crosswalks are too long, missing curb ramps or tactile markings while press buttons are positioned too high. These make it difficult for people with disabilities to proceed with their daily activities the same way people without disabilities do.
While mapping accessibility obstacles is challenging, systems like Project Sidewalk allow to virtually explore the streets of a city and label this type of information. Another way that these challenges could be overcome is by using aerial images or LIDAR data to locate crosswalks or measure the height of a sidewalk accurately.
Once the data is collected, an AI program can be designed to automatically detect these features in the whole city using computer vision techniques. The result could be a program that allows us to regularly measure accessibility in a cost-effective way and on a large scale.
Bicycle count prediction
The surveys conducted by the City of Amsterdam in partnership with World Enabled also found bicycles as another relevant accessibility bottleneck. This could be limited to a city like Amsterdam where bicycles are popular but could become an issue in cities where e-bikes have become all the rage lately. While temporary, bicycles are obstacles that can delay, block or be a hazard for people with disabilities.
Since they are dynamic, detecting bicycles is harder but historical counts of the number of obstacles (or bicycles) on the sidewalk can be used to predict how likely it is to find such an obstacle on the sidewalk. For example, the proximity of a bar would explain the presence of bikes during the evening while a school would have more bikes parked during the day.
With this prediction algorithm, the dataset can be used by city officials to help identify busy areas. They can then make informed decisions such as the addition of more bicycle parking in that area. Or the same data can be used to add an additional layer of information to existing route planning systems, allowing for a personalised route that avoids these temporary obstacles.
Venue accessibility assessment
This is a problem that can be very painful for people with disabilities, especially wheelchair users. The accessibility information about public venues is often entered manually and there is often information missing for plenty of areas in the city. As a result, people end up learning about accessibility issues only after going to a venue.
The survey found that some platforms such as Wheelmap can find information about wheelchair ramps and accessible toilets. While it is a great tool, it is limited by users entering the information. Street view images are another example where you can look for a certain venue and see if the door is wide enough. With computer vision techniques, all of these tasks could be replaced and made more efficient.
Using existing images, the City of Amsterdam aims to build certain accessibility guidelines and offer the ratings to places automatically. These ratings would allow visitors to quickly identify which facilities are offered for a large number of venues. At the same time, the city can also monitor which venues are accessible and see if it improves over the years.