A Non-Intrusive Drowsiness Related Accident Prediction Model Based on D-S Evidence Theory

In this paper, a non-intrusive accident prediction model based on the information fusion technique of Dempster-Shafer (D-S) evidence theory is investigated, which can infer the belief probability of the drowsiness related accidents by integrating information from driver´s eyes and driving performance. Firstly, to acquire the relevant physiological signals and driving performance data used in this proposed model, a driving simulator experiment is introduced. Secondly, three features reflecting driver drowsiness are quantitatively extracted, including gaze direction changing, blinking duration and standard lateral deviation. Finally, D-S evidence theory is applied to fuse these features. The inference results can predict over 70% drowsiness related accidents and therefore demonstrate the utility of this proposed model.