Kalman Filter-based Real-Time Traffic State Estimation and Prediction using Vehicle Probe Data

This paper presents a bi-level Kalman filter methodology for real-time traffic state estimation and short-term prediction at signalized intersections. At the upper level, turning movements are estimated using probe vehicle and upstream detector data. At the lower level, upstream approach density and queue sizes are estimated. This approach is validated with drone-collected data from a four-legged signalized intersection in Orlando, Florida. Compared to the baseline method that relies solely on probe vehicle data, the bi-level approach significantly enhances traffic state estimation and prediction accuracy. Specifically, the upper-level turning movement estimation shows a standard deviation improvement of up to 50% over the baseline. Additionally, the method provides predictions with a minimal standard deviation of 92.8 veh/hr at a 5% market penetration level. The lower level improves queue size estimation by up to 32.8% and traffic density estimation by up to 18.5%. These results demonstrate the approach's effectiveness and readiness for real-time application in traffic signal control systems.