Comprehensive Drowsiness Level Detection Model Combining Multimodal Information

Abstract

This paper presents a drowsiness detection model that is capable of sensing the entire range of stages of drowsiness, from weak to strong. The key assumption underlying our approach is that the sitting posture-related index can indicate weak drowsiness that drivers themselves do not notice. We first determined the sensitivity of the posture index and conventional indices for the stages of drowsiness. Then, we designed a drowsiness detection model combining several indices sensitive to weak drowsiness and to strong drowsiness, to cover all drowsiness stages. Subsequently, the model was trained and evaluated on a dataset comprised of data collected from approximately 50 drivers in simulated driving experiments. The results indicated that posture information improved the accuracy of weak drowsiness detection, and our proposed model using the driver's blink and posture information covered all stages of drowsiness (F1-score 53.6%, root mean square error 0.620). Future applications of this model include not only warning systems for dangerously drowsy drivers but also systems which can take action before their drivers become drowsy. Since measuring the information requires no restrictive equipment such as on-body electrodes, the model presented here based on blink and posture information can be used in several practical applications.

Eye blink detection using CNN to detect drowsiness level in drivers for road safety

“…At first, the sensitivity of the sitting position and conventional indices are determined and a method that can detect the drowsiness was projected by combining the various indices related to the sensitivity of the feeble and the robust drowsiness. The projected method improved the accuracy of the feeble drowsiness detection with an RMSE of 62% [26].…”

Research on Face Local Attribute Detection Method Based on Improved SSD Network Structure

“…Different levels of pyramid structure are used to realize the effective detection of different targets. Because the bottom layer of the model has too small requirements for detection targets, the shallow layer of the network model can hardly obtain smaller target features, which limits the application of SSD based network structure model in small target recognition [16,17].…”

Early Identification and Detection of Driver Drowsiness by Hybrid Machine Learning

“…Sunagawa et al [7] discussed the actualized drowsiness location framework by utilizing EOG information. The first distinguished the eye squinted from the recorded EOG information and removed the eye top development parameters as highlights to be characterized utilizing Support Vector Machines (SVM).…”

“…𝐺 ∅ (𝑋, 𝑌) = ∅ 𝑒 ( ) ∅ (7) When the scale factor is given, the differential channel then determines the edge's magnitude and introduction. Edge data of various scales is used to obtain the final edge image.…”