Single-Frame Machine Learning Approach to Predict Vehicular Yielding Intention While Approaching a Pedestrian Crosswalk

Abstract

Pedestrian safety at uncontrolled crosswalks is a global concern, especially in India, where uncontrolled crosswalks contribute to high fatality rates. Predicting driver yielding intentions is crucial for improving safety, yet research in this area is limited. This study introduces a novel single-frame analysis using machine learning models and drone-captured high-resolution trajectory data to predict yielding intentions. The random forest model demonstrated superior performance, achieving an accuracy of 84%, with a precision of 83% and recall of 84%, outperforming other models in robustness and predictive reliability. Feature importance analysis identified vehicle speed, vehicle acceleration, and pedestrian speed as the most significant factors influencing yielding behavior. The innovation of this study lies in its application of single-frame analysis, a departure from traditional sequential models. This approach enables rapid, real-time predictions essential for immediate pedestrian safety interventions, reducing computational complexity, and sidestepping issues common to data-heavy sequential methods. The effectiveness of single-frame analysis could revolutionize pedestrian safety measures by integrating these predictive capabilities into advanced vehicular technologies and dynamic urban infrastructure. Potential applications include smart crosswalks equipped with sensors and dynamic digital signage that utilize model predictions to improve real-time pedestrian safety. Additionally, traffic light systems and personal safety devices, such as smartphones or wearables, could be optimized to receive direct alerts, significantly enhancing pedestrian decision-making and safety. These applications can provide real-time alerts, significantly reducing the risk of pedestrian-vehicle collisions and enabling safer navigation for vulnerable road users.