Predicting Positions and Velocities of Surrounding Vehicles using Deep Neural Networks

Abstract

Prediction of surrounding vehicles motion is a basic feature of the most advanced driver assistance systems (ADAS). In this paper, we present prediction of positions and velocities of surrounding vehicles using deep neural networks (DNNs). Three different DNN architectures are designed and explored: feed-forward, recurrent, and hybrid. Training and validation data are generated using IPG Carmaker simulation environment. The reliability and accuracy of prediction models under simulated highway conditions and variable number of input-output time steps have been examined. Hybrid DNN showed better performance compared to feed-forward and recurrent neural networks.