Despite significant advances in information technology, issues related to social exclusion
persist, where a large part of the world population does not have access to digital content due
to a series of economic, social and educational factors. Some of the people excluded from the
digital universe are those with upper limbs impairments, which are normally required to
manipulate certain computational devices such as the keyboard. Therefore, digital inclusion
efforts have gradually become a social consensus as it involves integrating people with
technology and exploring effective methods to ensure the development of applications
accessible to all. Different adapted interfaces are being developed to replace conventional
peripherals. Many of these proposals are based on eye tracking using computer vision
techniques, which conveniently provides distance interaction, without the need for physical
contact with the device. Based on this context, the primary purpose of this research was to
implement a low-cost computer technology aimed for people with limited mobility, which
allows the control of the arrow keys by tracking the movement of the eyes in four directions
of interest (up, down, left and right). The system was developed using the C++ programming
language with the support of pre-implemented algorithms from the OpenCV library, and
focused on the training and application of the Haar Cascade classification method for the
localization of the eye region and in the usage of specific computer vision and digital image
processing techniques to determine the direction of the gaze based on the recognition of the
whiteness of the sclera. The results obtained in the experimental phase reveal that the
proposed algorithm demonstrates sufficient potential to enable the real-time use of the system
in an appropriate and functional manner. The model indicated satisfactory overall
performance in the eye detection by achieving a true positive rate of 82%, accuracy of 87%,
sensitivity of 82% and F1 score of 84% and managed to recognize the eye movements in the
four directions of interest, with an acceptable response time for running some computing
applications.
|
