Tracking Vehicles Equipped with Dedicated Short-Range Communication at Traffic Intersections

Abstract

In the near future, the traffic stream will contain both connected and autonomous vehicles with Dedicated Short-Range Communication (DSRC) vehicle-to-infrastructure (V2I) capabilities. With these new technologies, it will become possible to optimize the performance of traffic intersections so that wasted time at red lights and carbon emissions are minimized. Sensors, such as Doppler radar and traffic cameras, can use the data received at Road-Side Units (RSUs) from DSRC-equipped vehicles to assist with tracking and classifying all of the traffic approaching an intersection. In order to fuse information between multiple sensors, each sensor at the traffic intersection needs to compute the uncertainty about its estimate of the state of every vehicle it is tracking. In this work, we evaluate different tracking filters for their ability to estimate the state of a vehicle approaching a traffic intersection based on GPS data received over DSRC. We ran experiments with a vehicle equipped with a Cohda Wireless Mk5 On-Board Unit (OBU) and a high-precision GPS sensor to generate ground-truth data. We present a comparison of the performance of a linear Kalman filter, extended Kalman filter, and particle filter configured with different kinematics models. The effects of measurement bias in the GPS data in DSRC messages is also explored; we observe that without any bias estimation, the performance of the track filters degrades significantly.