Intersection SPaT Estimation by means of Single-Source Connected Vehicle Data

Abstract

Current traffic management systems in urban networks require real-time estimation of the traffic states. With the development of in-vehicle and communication technologies, connected vehicle data has emerged as a new data source for traffic measurement and estimation. In this work, a machine learning-based methodology for signal phase and timing information (SPaT) which is highly valuable for many applications such as green light optimal advisory systems and real-time vehicle navigation is proposed. The proposed methodology utilizes data from connected vehicles travelling within urban signalized links to estimate the queue tail location, vehicle accumulation, and subsequently, link outflow. Based on the produced high-resolution outflow estimates and data from crossing connected vehicles, SPaT information is estimated via correlation analysis and a machine learning approach. The main contribution is that the single-source proposed approach relies merely on connected vehicle data and requires neither prior information such as intersection cycle time nor data from other sources such as conventional traffic measuring tools. A sample four-leg intersection where each link comprises different number of lanes and experiences different traffic condition is considered as a testbed. The validation of the developed approach has been undertaken by comparing the produced estimates with realistic micro-simulation results as ground truth, and the achieved simulation results are promising even at low penetration rates of connected vehicles.