Visual Focus of Attention Analysis Using Convolutional Neural Networks

01 Jun 2020
  • # AI
  • # CNN
  • # TensorFlow
  • # TensorFlowJS
  • # React
  • # WebRTC

[Get Official Documentation here]

Visual Focus of Attention (VFOA) is what a person’s attention is on, as determined by the direction of their gaze, or more plainly, what a person is looking at.

VFOA analysis can be used to estimate whether a learner’s attention is on his instructor, whether an online test taker’s attention is on their screen or (suspiciously) not, whether an advertisement is attracting and maintaining people’s attention, whom someone is speaking to in a social gathering, and many other things.

Computerizing VFOA analysis would have benefits such as immediacy (so that the instructor is alerted, in real-time, when a learner’s attention has wandered), larger scale (so that it hardly matters how many learners are in the classroom or even how many classes the instructor is teaching at a time, such as in multi-site lectures) and, possibly, higher accuracy.

This project was an attempt to computerize VFOA analysis using a Convolutional Neural Network (CNN) model, for use in the context of online exam proctoring/invigilation. The VFOA analysis simply involved checking whether an online test taker’s VFOA was on their screen (meaning they probably aren’t cheating) or not (they might be cheating.)

The project also included a bare-bones web-based proctoring system for demonstrating the model.

See related project.