Squint Eye Detection using Machine Learning and Computer Vision

Strabismus, commonly referred to as squint eye, is an ocular disease that entails both eyes not being able to look at the same point at the same time; one of the eyes moves either inwards, outwards, upwards or downwards. If strabismus is not detected early enough, it results in amblyopia, which causes permanent vision loss and cannot be restored after the maturity of the vision system. Existing methods to diagnose the disease involve ophthalmologists, thus requiring expensive tools, making it unaffordable in rural India. This research article presents an automated near real-time squint eye detection system based on MediaPipe Face Mesh, OpenCV, Python programming language, and Random Forest algorithm, trained with 16,654 eye images covering five categories – normal, esotropia, exotropia, hypertropia, and hypotropia. The method uses iris landmarks to calculate deviations of angle ratios between eye corners and irises, and makes use of hybrid decision-making with geometry and machine learning. The model attained accuracy of 91.4%, weighted F1-score of 91% and macro average F1-score of 83% on 3,331 images of test dataset. It works on regular laptops without GPU, produces PDF reports and is integrated into a web application.