Gaze Zone Classification for Driving Studies Using YOLOv8 Image Classification

Area of Science:

• Computer Vision
• Road Safety
• Human-Computer Interaction

Background:

• Gaze zone detection is crucial for road safety research, identifying driver attention areas.
• Existing methods for automatic gaze zone annotation can be complex and require pre-processing.
• Developing an accurate, user-friendly system is essential for practical application.

Purpose of the Study:

• To develop and validate an automated gaze zone detection system using YOLOv8 for road safety research.
• To assess the accuracy of YOLOv8 for gaze zone classification across different datasets and conditions.
• To provide user-friendly applications for data collection and model training.

Main Methods:

• Utilized YOLOv8 for image classification of driver gaze zones.
• Tested the system on existing and newly collected datasets with varying numbers of gaze zones (9, 10, and 12).
• Trained YOLOv8 models specifically for driver demographics and conditions (e.g., glasses, sunglasses).

Main Results:

• Achieved near-perfect accuracy in gaze zone detection without image pre-processing.
• Demonstrated high performance when YOLOv8 models were tailored to specific drivers and driving conditions.
• Developed companion apps for image collection and YOLOv8 model training/application.

Conclusions:

• YOLOv8 offers a highly accurate and efficient solution for automated gaze zone detection in road safety.
• The system's accuracy is dependent on training data reflecting specific drivers and conditions.
• Further research is needed to evaluate performance in diverse, real-world driving scenarios.