Evaluation of 1D and 2D Deep Convolutional Neural Networks for Driving Event Recognition

Area of Science:

• Automotive Engineering
• Machine Learning
• Signal Processing

Background:

• Driver behavior recognition is crucial for safety, insurance, and vehicle management.
• Inertial Measurement Unit (IMU) sensors in smartphones can capture driving event data.
• Traditional methods often rely on manual feature extraction, which can be complex.

Purpose of the Study:

• To investigate the effectiveness of supervised deep learning models, specifically Convolutional Neural Networks (CNNs), for driving event classification.
• To assess end-to-end CNN models that integrate feature extraction and classification.
• To classify driving events from smartphone IMU sensor data (linear acceleration and angular velocity).

Main Methods:

• Utilized supervised deep learning models, including 1D and 2D Convolutional Neural Networks (CNNs).
• Processed linear acceleration and angular velocity signals from smartphone IMU sensors.
• Classified driving events such as accelerating, braking, lane changing, and turning.

Main Results:

• The best performing CNN model achieved an accuracy of 82.40%.
• Macro- and micro-average F1 scores reached 75.36% and 82.40%, respectively.
• Demonstrated high performance in classifying various driving events.

Conclusions:

• Deep learning models, particularly CNNs, are effective for classifying driving events using smartphone IMU data.
• End-to-end CNN models offer a promising approach for driver behavior recognition.
• Accurate driving event classification can support the identification of aggressive and non-aggressive driving patterns.