MediaPipe Proctoring Toolkit

This repository presents face analytic algorithms for proctoring purposes based on MediaPipe framework.

Features

• Eye Blink Detection
• Face Orientation (2 DoF, horizontal and vertical)
• Facial Activity
• Face Movement
• Facial embeddings
• Facial expression

Requirement

• Mediapipe v0.8.10.2 (Simply checkout on this commit)

Installation

To install the toolkit, you need to first install mediapipe and the checkout to the specific version, mentioned previously.

git clone -n https://github.com/google/mediapipe.git
cd mediapipe
git checkout 63e679d9

Then, you can clone this repository under mediapipe root directory.

git clone https://github.com/sawthiha/mp_proctor.git

Demo Application

To run the demo app, you have to build it first:

bazel build -c opt --define MEDIAPIPE_DISABLE_GPU=1 mp_proctor:demo_app

If there is no build error, you can run the application using the following command.

GLOG_logtostderr=1 bazel-bin/mp_proctor/demo_app \                     
  --calculator_graph_config_file=mp_proctor/graphs/proctor_cpu.pbtxt

Troubleshooting

Build errors

fatal error: 'opencv2/core/version.hpp' file not found

This error occurs when OpenCV installation or config is not detected. If you are on Linux, you can solve this by running the setup_opencv.sh provided by Mediapipe. You can find it in the root directory of MediaPipe.

chmod +x setup_opencv.sh
./setup_opencv.sh

Related Research Papers

Efficient Online Engagement Analytics Algorithm Toolkit That Can Run on Edge

Research Materials can be found on this repository.

@Article{a16020086,
AUTHOR = {Thiha, Saw and Rajasekera, Jay},
TITLE = {Efficient Online Engagement Analytics Algorithm Toolkit That Can Run on Edge},
JOURNAL = {Algorithms},
VOLUME = {16},
YEAR = {2023},
NUMBER = {2},
ARTICLE-NUMBER = {86},
URL = {https://www.mdpi.com/1999-4893/16/2/86},
ISSN = {1999-4893},
DOI = {10.3390/a16020086}
}