A Computer Vision-Based Eye-Tracking System Toward an Eye-Controlled Powered Wheelchair

This paper presents the design of an advanced wheelchair system for patients with Amyotrophic Lateral Sclerosis (ALS), which controlled by eye movements in a natural environment. The system comprises an electric wheelchair, a vision system, and a primary control system. The control system captures the user's eye image via a camera, then employs deep learning and an attention mechanism to determine the direction of eye movement. A lightweight eye-movement recognition model (YOLOv5-small) is integrated into an embedded AI controller. The experimental results demonstrate a 93.5% accuracy in recognizing eye-movement direction, and a maximum wheelchair movement speed of 0.5 m/s.