A Vision-Based Approach for the Diagnosis of Digital Asthenopia

With the advent of the pandemic, a sea change has been observed with working professionals having seen their work transition online where dependence on electronic devices have increased with online meetings. The same is true for students as well as younger children who cannot go outside to occupy themselves. Prolonged exposure leaves them susceptible to Computer Vision Syndrome or Digital Eye Strain also termed as Digital Asthenopia. The proposed system recognizes the need to combat digital eye strain and diagnose it using a Vision Based approach by receiving footage of an individual while they are occupied with the use of an electronic device facing a screen and notify them. This is done by the detection of the parameters of blinking rate, redness of the eyes and dryness of the eyes. The classification model that is developed analyzes footage of the human eye and identifies whether or not the individual is suffering from digital eye strain. It is done through a YOLO based Object Detection model where the model was trained over a self-built dataset of stills of the human eye with samples marked as healthy or red eyes and healthy or dry eyes. The calculation of blinking rate and ascertainment of whether it is above a typical rate or not is done by the calculation of Eye Aspect Ratio (EAR) using a dlib model to identify facial landmarks. The dataset collected is initially preprocessed by annotation following which augmentation of the images was done with horizontal and vertical splits as well as 90-degree rotations. The results obtained were the successful identification of blinking rate as well as the accuracy and success of the object detection models that were developed represented by a Mean Average Precision (MAP) of 0.85 and 0.75 for redness and dryness respectively.