Our work builds on the idea to formulate a pipeline which can detect levels of activeness in real-time, using a single RGB image of a target person. It expands the aim to create a generalized solution which works under any/most configurations, be it in an interview, online class, security surveillance, et cetera.
We introduce a novel pose encoding technique, which encodes the 2-Dimensional keypoints extracted using Human Pose Estimation (HPE) algorithm.
Our alerting mechanism is wrapped around the whole approach; it provides a solution to inhibit low-activeness by sending notification alerts to individuals involved.

The pipeline can be run on a CPU, as well as on a dedicated GPU. We recommend using a dedicated GPU to achieve our framerate of ~35fps with a single Nvidia GeForce GTX 1650 graphics card.

1. Anaconda
2. Python3
3. PyTorch
4. scikit-learn
5. OpenCV

NOTE: Dependencies can either be installed individually, or a GPU enabled Anaconda environment can be created from the environment file using the following instructions:

conda env create -f ActiveNet_Environment.yml
conda activate ActiveNet
python demo.py --source <filename or 0 for webcam>

NOTE: To run the demo on CPU, add extra flag --cpu to the above command.

1. Aitik Gupta
   ABV-IIITM, Gwalior
   [email protected]
2. Aadit Agarwal
   ABV-IIITM, Gwalior
   [email protected]