Traffic Pattern Analysis at Urban Intersections through Vehicle Detection in Aerial Imagery

Abstract. This paper explores the application of aerial image sequences to analyze vehicle flows at urban intersections, with the goal of generating data that can inform adaptive traffic signal timing and improve urban mobility in complex, interconnected road networks. Using deep learning techniques, we detect vehicles in aerial images taken at different times of the day and week at a given urban intersection. This approach allows us to infer vehicle density, identify queuing patterns, and analyze traffic light cycles. Such analysis can assess the robustness of signal timing under different traffic flows and the factors that influence intersection performance. In addition, the use of aerial imagery allows for the derivation of often inaccessible signal timing plans while providing extensive coverage to evaluate surrounding traffic conditions that may impact the target intersections. Ultimately, this research contributes to improved traffic management strategies and supports the development of smart city infrastructure.