Streetseek - Understanding public space engagement using deep learning & thermal imaging

In this paper, a platform for analysing public space engagement is described. This research focused on efforts to better understand the various ways people interact with the city environment, for example; the number of persons on a street, the average time spent, and topically - due to Covid-19, the physical distance maintained between people. A novel data collection method was used to capture imagery from several streets in a low-cost, scalable, and privacy ensuring fashion. Insights were captured in real-time over several months on a five-minute interval, for nine hours a day and seven days a week, across multiple cameras. These insights were generated through a novel CNN trained on thermal camera imagery - which maintained the individual's right to privacy by ensuring that no person was identifiable in the captured data-set. Finally, a SORT based tracking algorithm was used to measure interactions over time.