ActiveNet
- 1 minTable of Contents:
ActiveNet is a multi-stage mechanism to detect levels of activeness, with a single monocular image input of a target person.
In-Depth Details
Our paper, ActiveNet: A computer-vision based approach to determine lethargy consists of in-depth details, literature review, experimentation results, etc. It can be found on arXiv.
Implementation Details
Overview
Multi-stage flowchart of ActiveNet
Alert mechanism flowchart
Modules
Human Pose Estimation
Modified this implementation to interpolate 2 extra keypoints from ground-truth 17 keypoints, used their weights trained on MSCOCO dataset.
Pose Encoding
This module creates the angular encodings from bare keypoint annotations from HPE module.
Multi-Class Classification
Level of activeness assessed into 1 of 4 levels with a Decision-Tree based classifier, trained on a self-scraped dataset.
Alarm System
Low activeness alerts run out to users of a Slack Workspace.
Conclusion
Our paper has been accepted in ACM India Joint International Conference on Data Science and Management of Data, (CoDS-COMAD ‘21)!
The whole pipeline is open-sourced on GitHub.
Aitik Gupta
I go by @aitikgupta throughout the web! \o/