The problem Advanced Distributed Surveillance System solves

It is very difficult to utilize all the public surveillance capabilities of a big city and often time pre symptoms of crime or very big give away factors are missed due to the fact that not everyone can man public cameras or keep an eye on everyone physically. This leads to blind spots for the security bodies to functions and in turn, these blind spots are exploited by anti-social elements of the society. These blind spots can be avoided and there can be warnings given ahead of time of a crime by looking at pre symptoms of a crime (or situations that lead to crime) or there can also be sure sign warnings that prompt for quick action from the Police or any security body.

All this can be done by developing a distributed surveillance system that takes signals from two of the smartest beings on earth that are capable of detecting pre symptoms as well as sure sign signals of crime.

• Humans
• AI

We have tried to do just this by combining user warnings and feedbacks with our trained AI. The whole system if we let it scale can make neighbourhoods super safe and can also lead to a reduction in action to emergency time as this can prompt actions to be taken before an event occurs.

To sum it up, We are trying to be your friendly neighbourhood spider man but with a catch that we would be working for the security body and use our spidey sense to inform them about any anti-social activity that may be going on in their neighbourhood. This would make the city safer and the job of the police/security body a little bit easier(or so we hope).

Challenges we ran into

Building a prototype in just 48 hours is a really challenging task especially when you're trying to work with technologies that are not deterministic in nature (yes, including Machine Learning, Deep Learning & CSS).

These are a few challenges that we ran into :

• Annotating dataset that we have to use for TensorFlow object detection API for detection of guns and knives.
  • We got over it by eventually realizing that we will have to annotate by hand and since we don't have enough time we will try to make the best of however many images that we can annotate.
• Getting Tensorflow Object Detection API to work on our local systems and selecting a pre-trained model that would perform well on our task.
  • It was a difficult task indeed but hours of debugging and not giving up gave us some good luck and eventually we were able to identify and train a pre-trained model for our task.
• Creating Node Server, React Native App, React Dashboard, & an Object Detection App (Python) in 48 hours and developing an architecture where they can intercommunicate successfully creating a sane model for surveillance systems.
  • This is a challenge that we could only fix partially by thinking of some ideas and implement a subset given the time constraint. We haven't solved this challenge yet due to the nature of the idea being so open-ended, we hope to solve this as we scale up our idea and build something that can make a difference