The problem Protest Management solves

Nowadays, we can see many amiable protest turn to violent and aggressive one , leading to death of many lives.
The government and media hive the situation and try to turn the protest in the favour of the ruling party made it difficult to identify the real motive of the protest. As it is not possible to deploy large number of police personnels at different locations instantly thus our project will help in prioritising the situations and reacting accordingly.

It often becomes difficult for the law enforcement agencies to handle such a situation without casualities and are unable to find the real culprit behind the violent escalation of any protest.
So the basic idea behind our project is to analyze the current situation of the protest . This is achieved by monitoring the crowd using the feed of the cameras like CCTVs or Mobile Phones.

The data obtained from these input devices is then transmitted to our Servers deploying deep learning models . What our models actually do is to scan the video frame by frame and then it examines each frames using certain algorithms to generate a feedback. The feedback is received in the form of percentage . This percentage represents the level of violence during the protest.

The model proposed by us can be used by the law enforcerment agencies to monitor and control the situation of the protest (in case it becomes violent ) and also help them to flag out the real culprits behind the actions .

As this process is completely automated , it will ensure quick response and minimum casualties.

This output will also be available on a website which other people can access .
This will help them to take a safer route for their daily works and hence ensuring safety of the people .

Challenges we ran into

1. As there was less time for retraining models so to overcome this we used a pre-trained model for fast prototyping.
2. Another problem was that Different models work on different versions of the same library.
3. Availability of conda GPU was a big issue we faced so we used cloud interfaces like Google Colab and Kaggle.
4. There was also a problem with GPU during the use of the Heroku server.
5. How to forward link into cv2 input of video stream so to overcome this we used ngrok server.