Sensor-Based Classification of Primary and Secondary Car Driver Activities Using Convolutional Neural Networks

Area of Science:

• Neuroscience
• Computer Science
• Automotive Engineering

Background:

• Driver safety relies on situational awareness and adaptability.
• Existing research often focuses on driver behavior anomalies and cognitive monitoring.
• Recognizing basic driving activities is crucial for advanced driver-assistance systems.

Purpose of the Study:

• To develop a classifier for basic driving activities using electrooculographic (EOG) signals.
• To adapt methods from daily life activity recognition for driving contexts.
• To evaluate the performance of a 1D convolutional neural network (1D CNN) for this task.

Main Methods:

• Utilized electrooculographic (EOG) signals, which measure eye movements.
• Employed a one-dimensional convolutional neural network (1D CNN) for activity classification.
• Trained and tested the classifier on 16 primary and secondary driving activities.

Main Results:

• The overall classifier achieved 80% accuracy for 16 driving activities.
• Specific driving activities like crossing roads, parking, and roundabouts showed high accuracy (97.9%, 96.8%, 97.4%).
• Secondary driving activities achieved an F1 score of 0.99, outperforming primary activities (0.93-0.94).

Conclusions:

• The proposed EOG-based 1D CNN classifier is effective for recognizing basic driving activities.
• The system demonstrates high accuracy for specific maneuvers and secondary actions.
• This approach shows promise for integration into driver safety and monitoring systems.