SA-4

An intelligent monitoring application that significantly enhances the pre-existing surveillance system, by utilising "Image Captioning" using DL and Data analytics, for smart crime detection.

Comment

Inspiration

A plethora of surveillance devices is being used by the Defense Services for supervision and monitoring. However, most of them are manually operated at the cost of enormous amounts of time and manual labour.

What it does

We propose a Deep Learning Application that will be able to solve the above-mentioned problems.

  • Our application named ‘Cap-Bot’ is capable of running Image Captioning on multiple CCTV footage and storing the captions along with the camera number and the time of capture in a convenient log.
  • The file of saved captions can then be used to look up incidents from any instant of time just by entering a few keywords. The returned camera number and time slot can then be used to obtain the required CCTV footage.

How we built it?

We use a streamlit platform to take input for the captioning model which in turn produces a CSV file with the captions. Now, the CSV file is available for the user to download and further upload to our search interface for the user to intuitively pinpoint person and point of interest.

Advantages of the proposed Hack

  • Since our model relies on Deep Learning, the time can be reduced considerably as we are resorting to an automatic searching operation.
  • Manual Labour can also be eliminated by a notable portion as the task of both monitoring and maintenance will mainly be performed by the application.

Challenges we ran into

The major challenges that we ran into were:

  • Division of labour according to the specializations of team members, especially as the team consists of amateur ML enthusiasts.
  • The model had to be trained in a different way to include intricate objects and situations.
  • The generated CSV file was turning out to be really heavy and had to be processed to be computationally efficient.

Accomplishments that we're proud of

We are proud of the efficiency that our model is working on and the inference time we are able to achieve on a normal PC. Also, the kind of security we are able to promise with our application.

Built With

Share this project:

Updates