STUDY AND DETERMINATION OF EYE ASPECT RATIO USING DATA MINING

Abstract

This research paper studies the landmarks of the eye and analyses the value for Eye Aspect Ratio, which is the ratio of the average of Euclidean distance between pairs of landmarks in vertical position and the Euclidean distance between the horizontal landmarks of the eyes. This ratio has a value lying in a particular range. This ratio is of crucial importance in the blink detection using computer vision. Also this value will be used for other computer vision related tasks like behavior and facial expression detection. Our aim is to find the range of EAR value using which the results of detection will have optimal error in accuracy. An exploratory data analysis on data collected from real world and for various scenarios is conducted. A supervised machine learning model will also be trained and deployed for frame to frame prediction. Their combined result is demonstrated with visualization.