Non-traffic-induced vehicle stops and their impact on traffic congestion with drone footage

Abstract

This study investigates the impact of non-tra c-induced vehicle stops  such as pick-up/drop-o activities, illegal parking, and delivery operations  on urban tra c congestion. We propose a novel methodology for identifying these stops using high-resolution drone data and machine learning techniques, including Random Forest and XGBoost. Stops are detected from vehicle trajectories and classi ed as tra c-induced or non-tra c-induced based on contextual features, such as stop duration and location characteristics. While non-tra c-induced stops are predominantly observed in the rightmost lane  as expected  the results further provide a quantitative characterization of their temporal and spatial patterns, enabling the identi cation of critical hotspots and the estimation of their impact on overall tra c dynamics. In addition, we nd that short-duration non-tra c-induced stops increase lane-changing activity, circulating vehicles change lanes to avoid the stopped vehicle, which can lead to shockwave formation and reduced tra c ow e ciency.