I Spy With My Little Eyes: A Convolutional Deep Learning Approach to Web Eye Tracking

Eye-tracking is the study of eye movements, blinks, fixations, and is aiming to give insight into visual attention mechanisms. Being a common in marketing, usability research, as well as in cognitive science, there are well stablished methods for lab eye tracking, yet web eye tracking uses webcams of much lower quality. Web eye tracking can provide valuable information about users’ engagement with digital content from the comfort of their own home. This gives designers, developers, and researchers the chance to inform their decisions from data and optimize e.g. user experience while connecting to large and demographically diverse samples without the necessity for lab-level equipment. For web eye tracking, only limited tools exist that are accompanied with uncertainties which need to be addressed before using these tools for scientific research. Improving the quality of data collected via such channels is also part of this goal. The project aims to develop a reliable deep learning solution such as a convolutional neural network capable of predicting gaze x/y screen coordinates from the webcam video of users. The predictions of the proposed methods are compared to baselines models that use webcam data and to predictions made by the lab eye tracker.