Development of driver-state estimation algorithm based on Hybrid Bayesian Network

Abstract

In this study, we develop and evaluate an estimation algorithm of abnormal driving states (drowsiness, distraction, and workload) based on a Hybrid Bayesian Network (HBN) using multimodal information. The HBN algorithm is expected to increase transportation safety by combining merits of both the Bayesian Network and clustering algorithm. In addition, multimodal data efficacy analysis through human-in-the-loop experiments is used to enhance the performance of the driver-state estimation algorithm. Performance results obtained the lowest false alarm rate and fastest calculation speed. The false alarm rate decreased from 18.2 to 15.5%, whereas the calculation speed decreased by 4.35%.