A Real-Time Safety-Based Optimal Velocity Model

Modeling safety-critical driver behavior at signalized intersections needs to account for the driver’s planned decision process, where a executes plan avoid collision in multiple time steps. Such process can be embedded Optimal Velocity Model (OVM) that traditionally assumes drivers base their “mental intention” on distance gap only. We propose and evaluate data-driven OVM based real-time inference of roadside traffic video data. First, we extract vehicle trajectory data from footage through our advanced processing algorithm (VT-Lane) study site Blacksburg, VA, USA. Vehicles engaged car-following episodes are then identified within extracted trajectories database, time-to-collision (TTC) is calculated all instances. Then, analyze predict shape underlying TTC-based desired velocity function. A clustering approach used assess heterogeneity understand reasons behind outlying driving behaviors intersection design model accordingly. The results this assessment show calibrated replicate observed by capturing acceleration pattern with an error 20% lower than distance-based OVM.