Scientists identify detailed neighbourhood demographics from cars in Google Street View images


A report for the US National Academy of Sciences explains the methods used by a team of computer scientists to derive accurate, neighbourhood-level estimates of the racial, economic, and political characteristics of 200 US cities using the images collected by Google Street View in 2013 and 2014. The key element: the pictures captured of 22 million cars parked along or driving down those streets. The scientists trained a computer algorithm to recognise the make, model, and year of each automobile. The researchers then matched the vehicle data by zipcode (postal code) to the numbers on race, income, and education collected by the American Community Survey. A second algorithm looked for patterns in the types of vehicles typically owned in neighbourhood with specific characteristics. The scientists found many correlations between the cars found in a neighbourhood and its racial makeup, residents' level of educational attainment, and other characteristics. Finally, the scientists tested the algorithm by asking it to infer the demographics of the 85% of US zipcodes not included in the original test from just the car data. The result was surprisingly accurate, particularly at predicting the percentage of Asians, blacks, and whites in a given area. Given additional detail - by extracting features of homes, landscaping, and public areas - the researchers believe they could achieve greater accuracy.

Writer: Chistopher Ingraham
Publication: Washington Post

Related learning resources
Target Profile