Improvement of driver- state estimation algorithm using multi-modal information

This paper aims to improve the performance of drivers' state estimation algorithm using multi-modal information. While driving, drivers' abnormal states may increase the likelihood of an accident. Thus, detection of the precise states of drivers is a key factor of preventing car accidents. Driver drowsiness, distraction, and workload was assessed through human-in-the-loop experimental data including vehicle, video, voice and physiological information. Dynamic Bayesian Network (DBN) was applied to assess the drivers' state and integrate sensor data, whereas Hybrid Bayesian Network (HBN) was used previously. This paper shows the improved performance - DBS estimates drowsiness with 67.3% correct detection (n=4), visual distraction with 82.8% correct detection (n=16), cognitive distraction with 80.6% correct detection (n=16), and high workload with 86.6% correct detection (n=16). This research showed improved performance compared to our previous algorithm shown in Ryu et al. (2015).