Sensor-Based Extraction Approaches of In-Vehicle Information for Driver Behavior Analysis

Area of Science:

• Automotive Engineering
• Data Science
• Sensor Technology

Background:

• Modern vehicles collect standardized data via Controller Area Network (CAN) systems.
• Extracting proprietary vehicle data (e.g., brake, steering) for driver behavior analysis presents challenges.
• Existing methods require complex electronic control unit identifier analysis and data interpretation.

Purpose of the Study:

• To develop an automated system for extracting proprietary in-vehicle information from CAN frames.
• To correlate sensor data with desired information for accurate extraction.
• To enable detailed driver behavior analysis through enhanced data acquisition.

Main Methods:

• Vehicle driving status estimation using Inertial Measurement Unit (IMU) and Global Positioning System (GPS) data via threshold, random forest, and LSTM techniques.
• Segmentation of CAN frames based on estimated driving status.
• Scoring and selection of CAN frame segments using a distance matching technique for similarity assessment.

Main Results:

• Driving condition estimation accuracy achieved 84.20%.
• Proprietary in-vehicle information extraction accuracy reached 82.31%.
• The system demonstrated feasibility for automatic proprietary data extraction in real-world urban driving.

Conclusions:

• The proposed system effectively automates the extraction of critical proprietary vehicle data.
• Accurate driving status estimation is key to successful information extraction from CAN frames.
• This approach offers a viable solution for advanced driver behavior analysis and automotive research.