Implementation of a High-Accuracy Neural Network-Based Pupil Detection System for Real-Time and Real-World Applications

Gabriel Bonteanu ¹, Petronela Bonteanu ², Arcadie Cracan ¹

⁰Fundamentals of Electronics Department, "Gheorghe Asachi" Technical University of Iasi, 700050 Iasi, Romania.

Sensors (basel, Switzerland) +

|

April 27, 2024

View abstract on PubMed

Summary

This study introduces an AI-powered pupil detection system using slim neural networks for real-time applications. Achieving 96.29% accuracy at 5 pixels with 100 frames/s processing, it enhances assistive technology and driver safety systems.

Area Of Science

Computer Vision
Artificial Intelligence
Biomedical Engineering

Background

Accurate pupil detection is crucial for human-computer interaction and monitoring systems.
Existing methods often struggle with real-world conditions like variable lighting and diverse datasets.
The need for efficient, high-accuracy pupil detection in real-time applications is growing.

Purpose Of The Study

To implement and evaluate a novel artificial intelligence (AI)-based pupil detection system.
To achieve high accuracy and processing speed for real-world, real-time applications.
To demonstrate the system's generalizability across diverse eye image datasets.

Main Methods

Utilized slim-type neural networks with a parallel architecture for reduced complexity.
Trained and validated the system on approximately 40,000 eye images from 20 diverse databases.
Employed two independent classifiers to determine pupil center coordinates.

Main Results

Achieved a detection rate of 96.29% within a 5-pixel threshold.
Reported a standard deviation of 3.38 pixels for detection accuracy across all datasets.
Demonstrated a processing speed of 100 frames per second (fps).

Conclusions

The developed AI pupil detection system offers high accuracy and processing speed.
The system's robustness and generalizability make it suitable for variable lighting conditions.
Potential applications include assistive technology (eye typing), gaming, and automotive driver monitoring for safety.

Keywords:

artificial intelligence classifier neural networks pupil detection real-time