Gaussian Process Regression for Tail Vehicle Departure Time Prediction at Signalized Intersections Using UAV...

Area of Science:

• Traffic Engineering
• Data Science
• Transportation Systems

Background:

• Traditional vehicle departure time prediction methods struggle with traffic uncertainties and multi-lane complexities.
• Existing models often assume steady-state traffic flow, limiting their real-world applicability.

Purpose of the Study:

• To develop a novel method for predicting tail vehicle departure time using unmanned aerial vehicle (UAV) trajectory data.
• To optimize green light crossing windows and eco-driving trajectories for connected vehicles at signalized intersections.

Main Methods:

• Leveraging UAV trajectory data to create fleet state features.
• Applying Gaussian process regression for departure time prediction modeling.
• Validating the model with field-measured data from multi-lane signalized intersections.

Main Results:

• Departure time of tail vehicles follows a Gaussian process, validating the chosen regression method.
• The proposed method achieved an average 5.146% reduction in Mean Absolute Percentage Error (MAPE) compared to benchmark models.
• The model demonstrated good robustness, performing comparably to XGBoost.

Conclusions:

• Gaussian process regression is effective for modeling tail vehicle departure times at signalized intersections.
• The UAV-based approach offers improved accuracy for traffic flow optimization.
• Further validation with cross-site data is needed to generalize findings to unseen intersections.