Multi-modal Biometrics Based Implicit Driver Identification System Using Multi-TF Images of ECG and EMG.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Vehicle security is increasingly important, driving research into driver identification using bio-signals.
• Bio-signals like ECG and EMG can contain artifacts from the driving environment, reducing identification accuracy.
• Existing systems often fail to adequately process these artifacts, leading to suboptimal performance.

Purpose of the Study:

• To develop a robust driver identification system that overcomes the limitations of existing methods.
• To improve the accuracy of driver identification by effectively handling bio-signal artifacts.
• To propose a novel approach utilizing multi-TF image conversion and multi-stream CNN for ECG and EMG signals.

Main Methods:

• Preprocessing of electrocardiogram (ECG) and electromyogram (EMG) signals.
• Conversion of preprocessed signals into 2D spectrograms using multi-time-frequency (multi-TF) image techniques.
• Implementation of a multi-stream convolutional neural network (CNN) for driver identification.

Main Results:

• The proposed system achieved an average accuracy of 96.8% across all driving conditions.
• An F1 score of 0.973 was obtained, demonstrating high precision and recall.
• The system outperformed existing driver identification methods by over 1% in accuracy.

Conclusions:

• The developed driver identification system effectively utilizes ECG and EMG signals via multi-TF image conversion and multi-stream CNN.
• The proposed method significantly enhances identification accuracy by robustly handling bio-signal artifacts in real-world driving scenarios.
• This approach offers a promising solution for advanced in-vehicle security systems.