PGV: An Edge Operator Based Random Point Localization Algorithm for Eye Tracking System

Eye tracking is an important variable and offers benefits in Human-computer interaction (HCI) which is often complex and impractical in practice. Eye tracking implementation usually requires specialized devices, technology, or setup to work. On top of that, challenges like face obstructions like glasses or masks, differences in lighting, distance, and angles add another layer to the implementation difficulty. A novel random point detector algorithm called Pixel Gradient Value (PGV) is proposed which is an interest point detector-like algorithm to be used for eye-tracking purposes, which works by calculating the gradient value of a pixel in a given neighborhood by using an edge detection operator like Sobel edge detector and Canny edge detector as its basis. PGV is image-agnostic, it is capable of covering a wide amount of eye-tracking scenarios while running on a minimum setup. PGV is then used in a series of Fully Convolutional Network (FCN) machine learning system used for classification and point localization of the pupil center point. The proposed model is able to achieve higher results compared to the state-of-the-art result of em ax ≤ 0.025 value of 96.04, em ax ≤ 0.05 value of 98.71 and em ax ≤ 0.01 value of 99.19.