Dual-Cameras-Based Driver's Eye Gaze Tracking System with Non-Linear Gaze Point Refinement.

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Automotive Safety

Background:

• In-vehicle driver gaze tracking is crucial for monitoring attention.
• Existing offline models suffer from poor generalization and estimation bias in online predictions.
• This bias causes inaccurate gaze point localization.

Purpose of the Study:

• To propose a novel driver eye gaze tracking method using non-linear gaze point refinement.
• To eliminate estimation bias and implicitly fine-tune gaze points for improved accuracy.
• To enhance the generalization performance of online gaze prediction systems.

Main Methods:

• A two-camera monitoring system is employed.
• A two-stage gaze point clustering algorithm extracts representative gaze zones (forward and mirror).
• A non-linear gaze point re-mapping relationship is established. Unscented Kalman filter tracks driver status features.

Main Results:

• The non-linear gaze point refinement method outperforms previous gaze calibration and mapping techniques.
• Improved gaze estimation accuracy is demonstrated, even in cross-subject evaluations.
• The system effectively predicts driver attention.

Conclusions:

• The proposed method successfully addresses estimation bias in driver gaze tracking.
• Non-linear gaze point refinement offers a robust solution for real-time applications.
• This system enhances driver monitoring and attention prediction capabilities.