2017 Google

Google Maps

Computer vision uncovers predictors of physical urban change

Streetchange is a new way of measuring changes in the physical appearances of neighborhoods using a computer vision algorithm. We calculate Streetchange by algorithmically comparing Google Street View images of the same location captured in different years.

Physical urban change—the evolution of cities' public and private infrastructure—has been of interest to policymakers, as well as scholars in economics, sociology, and urban planning for decades. However, it has been difficult to measure urban change in ways that are quantitative, robust, and scalable. In the present work, we developed a computer vision method that allows us to quantify urban change at high spatial resolutions using image time-series from Google Street View. We used this method to compute urban change for more than 1.5 million street blocks from five major American cities.

We aggregated Streetchange data from street blocks at the census tract level for the five American cities (Baltimore, Boston, Detroit, New York, and Washington DC). In our paper, we linked Streetchange to demographic and economic characteristics of neighborhoods, in order to understand which neighborhoods experience physical change. We found that physical growth occurs in geographically and physically attractive neighborhoods with dense, highly-educated populations.

Learn more here:

Method and Results

We accessed more than 1.6 million Google Street View images of street blocks from Baltimore, Boston, Detroit, New York, and Washington DC, captured from the same locations in 2007 and 2014.

For all images, we calculated Streetscore, a metric for perceived safety of a streetscape. Streetscore reflects the combined perception of thousands of internet users. We computed Streetscore using a regression model based on two image features: GIST and texton maps, obtained from pixels of two object categories—ground (containing streets and sidewalks) and buildings.