Intent Inference of Driver Deceleration Behavior by Using Unscented Kalman Filter Integrated with Conventional Artificial Neural Network Model

Abstract

Early brake pedal operation and corresponding vehicle deceleration are crucial to mitigating rear-end collision risks. In this paper, a mathematical model, hereafter referred to as the deceleration intention inference system (DIIS), was developed to facilitate determining the intent inferences of driver deceleration behaviors. More specifically, a conventional neural network model was integrated into an unscented Kalman filter in an effort to describe the deceleration intentions that can be expected to occur a few seconds later. The numerical examples provided herein show that our proposed model is capable of inferring driver intentions more precisely than the conventional approach.