STUDY ON DRIVER YIELDING PATTERNS AT PEDESTRIAN CROSSWALKS

Abstract

This study is carried out in two main parts. First, it introduces a novel safety measure, Heavy Braking Duration (HBD), to understand driver behavior at unsignalized pedestrian crosswalks more comprehensively. We analyze video data from an unsignalized crosswalk in Bhubaneswar, India, focusing on factors such as vehicle type, speed at the onset of braking, distance from the crosswalk at braking onset, and maximum deceleration during braking. The study identifies HBD instances where drivers had to decelerate at more than 4.5 m/s² upon detecting a pedestrian. We utilize the Weibull Accelerated Failure Time (AFT) model to investigate the relationship between these variables and HBD, taking into account both 'censored' and 'uncensored' HBDs. Our findings highlight that rapid deceleration, safe distance from the crosswalk, and controlled vehicle speed significantly influence HBD. The study underscores the importance of HBD as a nuanced metric for assessing driver behavior at unsignalized crosswalks, with significant implications for traffic management and policy interventions. Additionally, in the second part, prediction of driver yielding decision at a pedestrian midblock crosswalk is carried out using video-based data considering vehicle, pedestrian, and traffic-related factors. The data collection site was located on NH334 in front of COER University, at a midblock pedestrian crosswalk using a DJI Mini 3 Pro (DJI RC) drone. For prediction, four machine learning models (Random Forest, GBoost, AdaBoost, XGBoost) are employed, with Random Forest achieving the highest evaluation metrics, including accuracy, precision, recall, F1 score, and AUC-ROC.