Detection of driver drowsiness level using a hybrid learning model based on ECG signals

Area of Science:

• Artificial Intelligence in Transportation
• Biomedical Signal Processing
• Machine Learning for Driver Monitoring

Background:

• Driver fatigue significantly impairs operating ability and vehicle control.
• Accurate classification of drowsiness levels during driving is challenging.
• Electrocardiogram (ECG) signals offer potential for monitoring driver physiological states.

Purpose of the Study:

• To propose an intelligent algorithm for classifying driver drowsiness levels.
• To develop a hybrid model integrating deep learning and traditional machine learning techniques for ECG signal analysis.
• To enhance driver safety through early detection of fatigue and drowsiness.

Main Methods:

• Two models were established: a deep learning model (model_1) for raw ECG prediction and a PCA-WKNN model (model_2) for heart rate variability analysis.
• A hybrid model, DBPW (DiCNN-BiLSTM and PCA-WKNN), was developed by combining and weighting the predictions of model_1 and model_2.
• The DBPW model's validity was confirmed using simulations on a public ECG database.

Main Results:

• The DBPW model demonstrated an average accuracy, sensitivity, and F1 score of 98.79% across multiple drivers.
• Recognition accuracy for drowsiness or fatigue states reached 99.33%.
• The hybrid approach effectively integrated different signal processing and machine learning techniques.

Conclusions:

• The proposed intelligent algorithm can effectively identify driver anomalies, specifically drowsiness.
• This technology offers a novel approach for developing advanced driver-assistance systems in intelligent vehicles.
• The findings contribute to improving road safety by enabling proactive driver monitoring.