Correlation Analysis of In-Vehicle Sensors Data and Driver Signals in Identifying Driving and Driver Behaviors

Area of Science:

• Automotive Engineering
• Sensor Technology
• Machine Learning in Transportation

Background:

• Modern vehicles utilize numerous sensors communicating via Controller Area Network (CAN) bus for performance monitoring.
• Existing vehicle sensors offer potential for novel applications like driver behavior analysis and enhanced safety features.
• Key driving interaction data, such as brake pedal excursion, is not consistently available across all vehicle models.

Purpose of the Study:

• To investigate the correlation between in-vehicle sensor data and driver-specific signals.
• To evaluate the importance of typically unavailable signals, like brake pedal data, for driver analysis.
• To assess the performance of machine learning classifiers in identifying driving modes using combined vehicle and driver data.

Main Methods:

• Acquisition of data from the CAN bus, externally instrumented brake pedal, and driver leg using inertial sensors and electromyography.
• Evaluation of different sensor subsets for analyzing driver behavior and identity.
• Application of three distinct classifiers to categorize driving modes based on collected data.

Main Results:

• Classification accuracy for driver behavior identification exceeded 0.93 and driver identification reached 0.96 when both vehicle and driver data were utilized.
• Accuracy for behavior identification remained above 0.80 even when excluding driver-specific signals.
• A strong correlation was observed between vehicle sensor data and driver-derived signals.

Conclusions:

• The integration of driver signals significantly improves the accuracy of identifying driving behavior and individual driver characteristics.
• Vehicle data alone can provide substantial insights into driving behavior, with potential for further accuracy improvements.
• Findings support the development of practical embedded systems for real-time driving mode analysis and data logging, beneficial for logistics.