Video based prediction of engagement

Title:
Video based prediction of engagement

About the project:
Prediction of user engagement from video content based on head pose and eye gaze orientation.

Environment:
Ubuntu 20.04
ROS
Python3

The system:
Video frames are captured from a camera and engagement prediction is displayed on top of the output video.

Work done so far:
ROS integration
Video processing

Integration:
I will make available 10 images (whole or subject face only) and expect an engagement prediction with a value between 0 and 1

Required work:
From a batch of 10 images:
1. Extract features:
- head pose (x, y, z)
- eye gaze orientation (x, y)
2. Using the features, build a model, train, evaluate, test and save for further use.

Extracting the features should be possible with available libraries (I can suggest a few), however the engagement model should be constructed from scratch.

This project is for research purposes, does not require testing or any other commercial approach and must be completed in 1 week or less.