Analysis of Vehicle Location Prediction Errors for Safety Applications in Cooperative-Intelligent Transportation Systems

Abstract

Cooperative-Intelligent Transportation System (C-ITS) safety applications depend on reliable location information timely exchanged by road users. Due to inter-vehicle communication delays and sampling frequency, there always exists a time gap between the state observation update time and safety decision time. Predicting the vehicle’s locations into a future time epoch common to both host and subject vehicles enables real-time collision detection. Current studies of vehicle positioning performance mostly focus on the accuracy and availability of vehicle navigation solutions at equal observation intervals. Location error propagation over the prediction time intervals and dependence on various factors is not much understood. In this paper, we analyzed how the accuracy of the location prediction degrades depending on prediction intervals and state estimate errors from the measurement updates. We adopted the Kalman Filter method to predict locations with two representative location data sets collected in real road environments. Results from a dual-frequency Global Navigation Satellite System (GNSS)/Real-time Kinematic (RTK) receiver show that the Root Mean Square Error (RMSE) of prediction locations grow from a few centimeters at the state updates to about 50 and 100 cm within the prediction intervals of 1 and 2 seconds, respectively. This implies that GNSS/RTK positioning capability is a prerequisite for C-ITS safety applications. The experimental results from a surveying-grade GNSS/Inertial Navigation System (INS) receiver show that the RMSE can remain within 10 cm for the prediction interval of 2 s. High-rate INS velocity measurements provide significant advantages in efficient control of the error growth of the predicted vehicle locations.