Using Deep Learning for Urban Pedestrian Counting
Edward Deleu, Stefan Elez, Ansh Gadodia, Kyra Macvaugh, Grace Zhao · 2021
Overcrowding in urban areas makes pedestrian travel slower and more dangerous: disease transmission and theft occur more frequently in crowds. This paper proposes a deep learning system designed to analyze traffic camera footage, measuring crowd densities at five intersections in Manhattan. The results were plotted on a mapping model in MATLAB. The system measured crowd densities in clear weather with an average accuracy of 75% to 83% during the daytime and 59% to 79 % during the nighttime. The results demonstrate the viability of a pedestrian counting system for monitoring foot traffic in urban settings.