The problem Watchman solves

_Do you know how long it takes for a crowd to turn into a mob?
A simple push can lead to a brawl, and a brawl can lead to death. _

During the 2007 Davis Cup rubber between Chile and Argentina held in Movistar Arena, the Chilean crowd booed the Argentinian team and pelted Argentinian players and officials with fruit, coins, and plastic chairs, causing numerous delays.
In public places, violent behaviors pose a serious threat to personal security and social stability. It is very important to automatically detect violence in video surveillance scenarios, for instance even in railway stations, gymnasiums, stadiums, dark and dingy street alleys, and psychiatric centers.

Our web app, Watchman, helps in the detection of unruly public behavior of crowds and individuals and alerts the concerned authorities, thus generating a means of timely action to prevent grievous injuries, accidents, and eventual death.

Challenges we ran into

Some of the challenges we ran into:

• Accuracy: Initially the accuracy obtained by our CNN-LSTM model was 80%. We built upon the AlexNet model by modifying the network architecture and were able to improve the validation accuracy up to 96%.
• Alert Mechanism: Initially, we ran into a dead-end while exploring our options on how to alert the police forces. We finally resorted to pywhatkit as a medium to alert authorities in the event of a violent outbreak.