Abstract:
Researchers from all over the world have recently become increasingly interested in biobased human machine interfaces (HMI) for the assistance of paralyzed people enabling them to
live an assistance free life. Among various approaches of designing a Human machine interface,
eye signals are considered the most appropriate type of input. In general, eye-tracking systems
assess a person's eyeball position and gaze direction and are classified into two approaches:
electrooculography-based and computer vision based. This research uses EOG, and computer
vision technique to predict which input method is more appropriate and user friendly for the
mobility of an electric wheelchair. EOG data is acquired for four different eye movements i.e.,
right, left, upward, downward using BIOPAC. Video based data set is acquired using a webcam
mounted at a fixed distance from the subject. EOG dataset is then processed and classified using
eleven different classifiers among which the Decision tree shows the highest accuracy and F1 score
i.e., 88.94 ± 13.82, 89.12 ± 13.58 respectively. The other data set of videos is processed using
computer vision. Deep learning algorithms are used to classify the results. Both systems mentioned
in this study have their own limitations. For EOG based system, the attachment of electrodes is a
must requirement. This causes irritation to the user and sometimes generates motion artifacts
which can be a source of hinderance for the motion of any HMI. For computer vision-based system,
camera is a must requirement. However, it can’t be used in dark rooms, outdoor; during night
times, wearing sunglasses and in similar other situations. For such situations, another alternative
is an infrared camera, but prolonged usage of such camera can damage the eye. Therefore, a hybrid
system should be developed which involves both techniques i.e, EOG and a camera which can
effectively drive any mobility assistive device.
