Implementation of a CNN-Based Eyebrow Movements Detection System for Autonomous Wheelchair Navigation in Tetraplegic Patients

Recent advances in machine vision have enabled seamless integration with a variety of assistive technologies, one such application being autonomous wheelchair navigation. People with tetraplegia experience severely limited mobility due to an inability to move their limbs voluntarily, which makes operating a wheelchair independently extremely challenging. This project aims to tackle this issue by creating a real-time, non-invasive autonomous wheelchair navigation commands system through eyebrow movement recognition. There are four distinct eyebrow states (neutral, left raised, right raised and both raised) that represent commands for stop, turn left, turn right and forward respectively. Convolutional Neural Networks (CNN) were trained to classify eyebrow movements using a custom dataset from eight participants that involves data acquisition, preprocessing, model training and evaluation. Two eyebrow classifier models were developed to perform the identification of the left and right eyebrow movements. Both eyebrow classifiers showed high accuracy above 98% when evaluated on a test set. With its straightforward Graphical User Interface (GUI) and directional commands, the integrated real-time system demonstrated decent real-time applicability with low latency. As a low-cost, non-invasive assistive technology, it is a promising system in helping Tetraplegic patients regain some degree of autonomy.