Streets are conduits of human activity. Despite their importance, studying street activity has been obscured by a lack of data on how people use them, with most approaches limited to studying a single point in time or small geographic areas.

This paper proposes a new framework to measure street activity in real time. Our framework leverages machine learning and computer vision to classify pedestrian activities and transportation modes using images collected from moving vehicles. We apply our methodology to measure street activity in Paris for five weeks. We provide activity maps for this period and show that streets vary dramatically in their capacity to support pedestrian activity and that these differences are highly persistent. Our proposed framework can be used to measure street activities in other contexts and cities, providing urban researchers with an approach to guide planning interventions, identify infrastructural deficiencies, and inform design policies that foster active streets.