US-based researchers have created an algorithm that enables an assessment of security as felt by a city’s residents. The system analyses Google Maps Street View images and models the results on to a town map.

Mapping Public Perceptions of Safety to Guide Urban Development

Safety and security in cities is a key priority for local authorities and is sure to come up as one of the topics whenever there is debate about using Big Data to predict citizen behaviour. L’Atelier has already reported on the way Paris-based :Snips draws on city-related data to build mathematical models that provide simulations of peak passenger flows on train networks. Other data experts have been studying tweets, using an algorithm designed to predict the likelihood and geolocation of criminal activity in cities. Now the Macro Connections and Camera Culture groups at the MIT Media Lab in the United States have been working together to develop StreetScore, an algorithm which enables an assessment of the level of safety and security of a given place from a Google Maps’ Street View image, giving a score of 1 to 10, with 10 meaning that people perceive safety levels there to be high.

Modelling the perception of safety in cities

The sources that StreetScore uses were put together for an earlier project in 2011 called Place Pulse. The inhabitants of New York City and Boston were quizzed about their perception of safety in their cities.  They were shown Street View images and asked to choose which ones depicted higher levels of safety in their eyes. Now that this data is available, the results can be modelled on to a high definition map of the city so as to visualise public perceptions of safety. The algorithm gives a score of 1 to 10 to the various photos taken in the city and coloured dots appear on the map, ranging from green (very safe) to red (not at all safe). Nikhil Naik, a Ph.D student working with the Camera Culture group who developed the algorithm explains that StreetScore can help to spot specific features of the urban built environment which tend to indicate risk or a perceived lack of safety in the eyes of the general public.

Adapting urban development to perceived safety levels

As the algorithm is fed from data provided by the city’s residents, the researchers are essentially using the crowdsourcing approach to identify, or at least predict what might be the less, or more, safe areas of a town. To take a simple example, a StreetScore-generated city map shows that people see areas outside the city centre, such as bridges and motorways, as less safe. Town planners might therefore wish to use the map as a basis for making some changes to the layout of their city. “How can you make the city safer by changing or adding features in the built environment?” This is the question Nikhil Naik set out to answer when he began to develop StreetScore. Unsurprisingly, areas with green spaces and trees are seen as more welcoming than multi-storey brick buildings. At the moment only a few thousand images of US cities have been analysed but in time drawing on people’s perceptions in this way might serve to predict the likelihood of a crime being committed at a given location. Place Pulse is now looking at 56 more cities in order to build a broader set of data.

By Eliane HONG