An Intelligent Eye Gaze–Driven E-Book Reader using MediaPipe and OpenCV

Abstract

The fast switching to digital documents in educational and professional life has raised the number of screen-based reading enormously, but it is accompanied by a notable decrease in concentration, constant-distraction, and the inability to continue reading. To solve these problems, the following paper gives a gaze-assisted web-based reading platform that can track the eye gaze of users and tell them the line that must be read and also allow them to navigate through the pages of a book using only double and triple blast gestures. The suggested system combines a computer vision interface with the web application layer, communicating on a lightweight Representational State Transfer (REST) interface. A MediaPipe FaceLandmarker model is used to obtain facial landmarks in order to come up with normalized gaze vectors that are made by the displacement of the iris and the estimation of the head-pose which are filtered by an 8-frame median filter before being sent to the server. At the same time, Eye Aspect Ratio (EAR) -based analyzes of blink occurrences are designed to identify gesture intent by passing gesture intent validations based on inter-blink interval and a rhythm gate to avoid false positives. The threaded Flask backend synchronizes the processed gaze and gesture cues, and integrates with a Portable Document Format (PDF) rendering module to provide per-line highlight updates and page-turn feedback. The deployed system has shown gaze-based line tracking and high confidence hands-free navigation through blink gesture recognition at a timeframes (temporal) of 550 ms and thus the viability of an eye-controlled, accessible digital reading interface.