How can we interpret and analyze urban safety and users’ perception of safety through images of public space?
During the DataTHINK Workshop, we delved into a critical analysis of the role of AI in addressing this issue, trying to recreate the processes and algorithmic thinking that take place behind Vision and Large Language Models.
First, by assessing a sample of StreetView images of Barcelona and Montreal, we identified elements that make a place look more or less safe. This showed us that while some scenes are trivial to define as safe or unsafe, some other ones are for subjective and are dependent on the context, time of the day, knowledge about the city, among other consideations.
For this reason, we changed our point of view, changing from external agents to users of the public space. By walking around Barcelona we were able to identify from our own experience what elements made us feel safer. From punctual tactical urbanism interventions in the neighborhood of Sant Antoni, to the pedestrianization of Consell de Cent Street are examples of how urban design can translate into safer conditions for pedestrians.
This assessment both as observers and users, demonstrate the type of factors that have an effect in urban phenomena such as safety and users’ perception of safety. These include elements, actors and conditions.
Nevertheless, there is an extra layer of complexity for understanding this issue. Safety is a very broad concept, that could be understood as
- Road safety, referring to feeling safe from other street actors (bikes, moving vehicles or parked vehicles)
- Personal safety, referring to feeling safe from other human beings (theft, harassment or physical violence)
- Physical safety, referring to feeling safe due to the design of the space (freedom of movement for disabled people, good state of the pavement)
- … among others
Also, a space that feels safe for a given individual might feel unsafe for another person of different nature. Therefore it is important to also consider variables such as:
- Age
- Ability
- Memory
- Ethnicity
- National Origin
- Gender
- Sexual Identity
How can AI interpret and analyze urban safety?
It is important that whatever role AI takes into this topic, it takes into consideration all the elements, actors and conditions that correlate with the feeling and perception of safety. But it is also important to differ between the different types of safety as well as the different points of view for which safety might be understood different. So when when framing questions that could be prompted for/by a computer vision model, it should iterate through the diversity of these variables to provide enough specificity to obtain accurate results.
