Real-Time Eye Movement-Controlled Wheelchair Using Image Processing and SSD Network for Enhanced Directional Classification

Abstract

The war that passed through the country left many injuries, resulting in limb amputations or bilateral or quadriplegic paralysis. Consequently, there became an urgent need to use wheelchairs, which led us to address the problem of automating a motorized wheelchair and enhancing the mobility of people with special needs. This research focuses on utilizing eye movement to control and steer a wheelchair using advanced image processing techniques.

In this work, image segmentation and the Hough Transform were employed to preprocess eye movement images, improving the precision of feature detection. These segmented images were then used to train a Single Shot Detector (SSD) neural network to classify eye positions into three categories: Right, Left, and Neutral. This classification helps control the wheelchair's movement by detecting the user's intent based on eye direction.

The motorized wheelchair was converted into an electric wheelchair using two DC motors with carefully calculated power to withstand the weight of the chair and the expected user. Two specially designed control loops were used to manage the motors, while the Viola-Jones algorithm was adopted for initial face and eye detection. The image processing pipeline was then enhanced by combining Hough Transform for feature detection and segmentation, followed by training the SSD network for classification.

A model and software interface were designed using the MATLAB environment. The system was tested on five individuals (three males and two females), aged between 30-45 years and weighing between 65-85 kg, under daylight conditions and on a level surface. Each participant performed the test at least three times. The success rate for detecting eye movement direction and steering the wheelchair was approximately 85% after incorporating the enhanced image processing techniques.