Unity Human Eye Model for Gaze Tracking with a Query-Driven Dynamic Vision Sensor.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Computer Vision

Background:

• Objective gaze pattern and eye movement measurement are crucial for neuroscience and neurological disease detection.
• Current wearable eye-tracking systems are limited by high power consumption of conventional cameras.
• Neuromorphic cameras, like the query-driven Dynamic Vision Sensor (qDVS), offer a low-power alternative by processing only intensity-change events.

Purpose of the Study:

• To evaluate the feasibility of using qDVS for on-body eye-tracking.
• To compare the performance of qDVS-based eye-tracking against conventional methods.
• To present an open-source quantitative framework for comparing eye-tracking algorithms.

Main Methods:

• A realistic human eye model was created in the Unity 3D graphics engine for simulation.
• Eye-tracking performance using simulated qDVS frames was compared to traditional ellipse pupil-fitting and deep learning models.
• The framework allowed controlled comparison under varying illumination conditions.

Main Results:

• qDVS-based eye-tracking achieved high accuracy: 93.2% for horizontal (pitch) and 93.1% for vertical (yaw) gaze movement.
• The system demonstrated feasibility across four different illumination conditions.
• The developed quantitative framework is open-source and extensible.

Conclusions:

• The qDVS shows significant potential for low-power, long-term mobile eye-tracking applications.
• This research validates a novel approach for evaluating and comparing eye-tracking technologies.
• The open-source framework facilitates further advancements in eye-tracking algorithm development and application.