As part of the SAMHAR-COVID-19 Hackathon
An intelligent system that -
- Sees and predicts number of people in front of it.
- Alerts if number of people surpass a set threshold.
Data on which the model is trained is from ShanghaiTech can be found here: https://drive.google.com/file/d/16dhJn7k4FWVwByRsQAEpl9lwjuV03jVI/view
Trained Model can be found here: https://drive.google.com/file/d/1W58cGbFr2SJvOmBoPu-quIjO5i9g4hFL/view?usp=sharing
Used the state-of-the-art CSRNet to train the model.
MAE recorded is 75%, which can be increased more after more Indian crowd related data, specific to the cause.
A Snapshot of resulting system
- Train on more specific data like video streams, to get better model's performance.
- Tune various other hyperparameters involved in the CSRNet.
- Deploy the trained classifier on to a drone or a surveillance camera for various implementations like aerosol/sanitizer spraying, sharing GPS location of area or playing police sirens.
- Ashutosh Agrahari
- Ankur Veer
- Anshuman Singh