Deep Transfer Learning Enabled Eye Gaze and Blink-Driven Support System for Paralysed Patients

This paper addresses the major communication and mobility challenges faced by paralysed patients mainly with Locked-in Syndrome (LIS) and Amyotrophic Lateral Sclerosis (ALS), which makes the patient paralysed and causes speech disorders, and thus limiting their independence to talk and move freely. Traditional methods of controlling wheelchairs require appropriate position and lighting conditions. This research present a novel system that utilises eye-gaze detection to enable these patients to control a wheelchair, enhancing their ability to move wherever needed and communicate effectively. The approach employs advanced object detection techniques, specifically the YOLOv8 model, combined with gaze detection ability through an integrated Inception-ResNet-v2 architecture. This allows the entire system to accurately identify gaze directions (straight, left, right, closed (for reverse)) and YOLOv8 identifies the eye states (open or closed) compared to the conventional tracking methods, which fail under poor lighting conditions and precise positioning constraints. The developed model has attained an accuracy of 86.3% in detecting open and closed eyes, and an accuracy of 90.2% in eye gaze tracking, enabling effective communication through blink detection and mobility control through gaze. This innovative solution aims to improve the quality of life for paralyzed patients by providing them with a reliable means of mobility and communication, empowering them to engage more fully with their surroundings.