Project

StreetScore

Camera Culture

Groups

StreetScore is a machine learning algorithm that predicts the perceived safety of a streetscape. StreetScore was trained using 2,920 images of streetscapes from New York and Boston and their rankings for perceived safety obtained from a crowdsourced survey. To predict an image's score, StreetScore decomposes this image into features and assigns the image a score based on the associations between features and scores learned from the training dataset. We use StreetScore to create a collection of map visualizations of perceived safety of street views from cities in the United States. StreetScore allows us to scale up the evaluation of streetscapes by several orders of magnitude when compared to a crowdsourced survey. StreetScore can empower research groups working on connecting urban perception with social and economic outcomes by providing high-resolution data on urban perception.

This website is a collection of map visualizations of perceived safety of street views from cities in the US as predicted by StreetScore. We will be releasing a map of perceived safety for a new city each week. The StreetScore algorithm was created by Nikhil Naik as part of a collaboration between the Macro Connections group and the Camera Culture group at MIT Media Lab. Jade Philipoom created the visualizations presented in the StreetScore website.

Please send your questions/comments to streetscore@media.mit.edu 

Publications

StreetScore - Predicting the Perceived Safety of One Million Streetscapes (pdf)

Nikhil Naik, Jade Philipoom, Ramesh Raskar and César A. Hidalgo.

CVPR Workshop on Web-scale Vision and Social Media

Accepted (2014)