Gaze Focalization System for Driving Applications Using OpenFace 2.0 Toolkit with NARMAX Algorithm in Accidental...

Area of Science:

• Computer Vision
• Human-Computer Interaction
• Automotive Safety

Background:

• Driver attention monitoring is crucial for advanced driver-assistance systems (ADAS) and autonomous driving.
• Current gaze estimation techniques are often intrusive and expensive, limiting real-world application.
• Existing datasets lack real-world accident scenarios, hindering robust system development.

Purpose of the Study:

• To develop a low-cost, non-intrusive camera-based gaze mapping system for driver attention analysis.
• To integrate the OpenFace 2.0 Toolkit with NARMAX modeling for accurate gaze estimation.
• To validate the system's performance on a challenging real-world accident database.

Main Methods:

• Utilized the OpenFace 2.0 Toolkit for facial landmark extraction and head pose estimation.
• Employed NARMAX (Nonlinear AutoRegressive Moving Average with eXogenous inputs) modeling to map gaze parameters to screen regions.
• Developed a heat map visualization of driver focalization.
• Validated the system using the DADA2000 database, comparing results with a previous linear model and a commercial eye-tracker.

Main Results:

• The proposed system achieved comparable results to an expensive eye-tracker.
• Demonstrated accurate visualization of driver focalization in real traffic scenes.
• Successfully mapped driver attention using low-cost, non-intrusive methods.

Conclusions:

• The developed system offers a viable, cost-effective solution for driver attention monitoring.
• This technology can be used to create valuable datasets for training and validating driver monitoring systems.
• Enhances safety potential for Level 3 and Level 4 automated driving systems by providing reliable gaze estimation.