Research on imaging method of driver's attention area based on deep neural network

Area of Science:

• * Intelligent Transportation Systems
• * Cognitive Ergonomics
• * Computer Vision and Machine Learning

Background:

• * Driver visual attention is crucial for intelligent driving decision-making and behavior analysis.
• * Existing driver intention recognition methods suffer from issues like interference from wearables, high false detection rates (e.g., with glasses or strong light), and unclear field-of-view extraction.
• * There is a need for non-intrusive and accurate methods to analyze driver visual attention.

Purpose of the Study:

• * To propose a novel method for imaging the driver's visual attention area.
• * To overcome the limitations of traditional driver intention recognition techniques.
• * To provide a theoretical basis for dynamic driving behavior analysis and traffic safety.

Main Methods:

• * Utilized driver's field-of-vision images from dash cams and vehicle driving state data (steering wheel angle, vehicle speed).
• * Employed interpretability methods of deep neural networks for attention imaging analysis.
• * Performed attention imaging analysis on a virtual driver model based on vehicle data to infer human driver's visual attention area.

Main Results:

• * The proposed method successfully images the driver's visual attention area.
• * Demonstrated the capability for reverse reasoning of driver intention during driving.
• * Validated the effectiveness of using deep neural network interpretability for this task.

Conclusions:

• * The developed method offers a non-intrusive approach to analyzing driver visual attention.
• * This technique provides a foundation for advanced dynamic driving behavior analysis.
• * Contributes to the development of safer intelligent driving systems and traffic safety research.