The problem Vision solves

Surveillance means safety.
Despite so much surveillance, the crime rates are increasing exponentially. India is toppping the charts for wrong reasons. The main reason for this is the inefficient way of surveillance and monitoring being used in our country. Several hundreds and thousands of CCTV cameras are being installed every day but only a select few have the monitoring access to them. The problem lies in the fact that inspecting several hundreds and thousands of videos is very laborous and time-intensive task.
Only after a crime is committed, hundreds and thousands of footages are inspected to find the crime scenes evidence.

Is there a way to minimise this? What if the crime/violence is reported in realtime using technology.
In this project we try to explore this possibility. Introducing to you our project VISION.

We have devised a complete software solution for safety and surveillance where we convert the conventional CCTVs from an evidence collection to a crime prevention and detection tool to ensure safety and security. This real-time crime detection technology is integrated with security systems and Desktop Application to get push-in notifications in case any suspicious activity is discovered.

Our solution captures and pre-processes the video feed and stores the frames of the video using OpenCV. The frames are then processed and classified using deep learning architectures implemented on Tensorflow to detect any suspicious activity. This result is then conveyed to the user/authority using either a Desktop Application (Electron) or a Web Platform using push-notifications.

Challenges we ran into

During this project, we ran into several technical problems. A few of the major ones were:
1. Finding the Dataset: Violence and Non-violence being a very subjective matter, it was very hard to find a suitable dataset for training the model. We had to use several datasets for this project, and also scrape from the internet to make a good enough model with reasonable accuracy.
2. Ease of Use: The project holds no utility if its hard to access. Thus we had to make the UI as simple as possible without compromising the functionality. For this project, we chose Electron for Desktop Development and Django for API Management over the Internet.
3. Tensorflow Model: Configuring tensorflow to actually work with out framework proved to be avery difficult task as the frames were being generated in real time and had to be forwarded to a web page continuously for smooth experince and faster performnace.