AI & ML for Safety: Analyzing Connected Vehicle Data for Turning Movement Counts at Intersections

Abstract: The advent of Connected Vehicle Data (CVD) has opened new opportunities for improving traffic safety assessments by offering detailed, high-frequency data on vehicular movements such as speed, direction, and braking events. This research leverages artificial intelligence (AI) and machine learning (ML) techniques to analyze CVD and accurately estimate turning movement counts (TMC) at intersections, essential for intersection design, signal timing, and improving road safety. Traditional methods, including manual counting and video-based analysis, have limitations in scope and accuracy, whereas CVD provides continuous, granular insights that enhance modern traffic analysis.
This study explores three innovative approaches for estimating TMC using CVD: geofencing, the Open Source Routing Machine (OSRM), and a data-driven method. The geofencing method involves placing virtual boundaries around intersections to detect vehicle movements based on their entry and exit across these boundaries. This approach is simple and efficient, but it may face limitations in more complex intersections where vehicles deviate from standard paths, leading to potential misclassifications of movement.
The OSRM method uses a high-performance routing engine to match GPS trajectories to the road network, ensuring accurate vehicle path mapping even when GPS data is noisy or imprecise. While this method offers higher precision for large-scale datasets, it requires considerable computational resources to process each trajectory. It may also struggle with complex maneuvers, such as U-turns, which could introduce inaccuracies in turning movement detection.
The data-driven approach estimates TMC by filtering out anomalies based on vehicle heading angles at entry and exit points. This method effectively removes outliers but requires complete trajectory data and precise angle measurement for accuracy. It may face scalability challenges, especially when dealing with large datasets, due to the need to process multiple trajectory points to avoid misclassifications.
While each method presents unique strengths and challenges, the integration of AI/ML with CVD offers transformative potential for traffic safety assessment. CVD’s high-resolution, continuous data improves the accuracy and timeliness of turning movement estimates, supporting better infrastructure design and traffic management. However, challenges such as CVD’s bias toward newer vehicles and its exclusion of vulnerable road users (VRUs) must be addressed to ensure comprehensive and equitable traffic analysis.
In conclusion, this study demonstrates that AI/ML techniques applied to CVD can significantly enhance traffic safety assessments. By addressing the limitations of traditional methods, these approaches offer scalable, data-driven alternatives for analyzing traffic dynamics at intersections. Future research should focus on refining these methods and integrating additional data sources, such as smartphone GPS data or roadside sensors, to provide a more holistic view of traffic behavior and ensure safer, more efficient transportation systems.