Inspiration

Public safety has been a concern for many people in our communities and can always be improved. As members of our community, we want to take initiative towards enhancing the safety of our people. We noticed the current form of notifying law enforcements could be improved significantly enough to save more lives in situations involving improper use of guns.

What it does / What we built

We built a platform that can significantly reduce the time it takes for authorities/law enforcement to be informed of a dangerous situation involving a gun. This platform allows existing security cameras to detect guns in real-time. Once a gun is detected, the platform does following:

  1. Store and Map the location of where the gun was detected
  2. Law enforcement will be notified of the location, video footage, and images of the suspect (via Web App)
  3. SMS (containing a photo of the suspect and situation) notifications are sent to people within the vicinity

How we built it

-----Machine Learning-----

-Gathering Data: Collected 300+ images, including annotations

-Training Data: Preprocessed data to optimize for training

-Darkflow (Tensorflow): Trained custom object detector to detect guns in realtime

-----Cloud Services-----

-Google Cloud Platform: Google Video Intelligence API and Google Vision API for obtaining faces of suspects

-Amazon Web Services: AWS SNS service was used to send SMS notifications when gun is detected

-----Web Application Platform-----

-HTML/CSS

-Javascript

-Bootstrap

-Google Maps API

Challenges we ran into

-Acquiring large dataset of images for our project

-Finding the optimal learning rate, # of epochs, and model threshold

-Detecting a gun in multiple videos with different settings

-Detecting a gun in realtime without frame rate drop

-Training the model within the time constraint of a hackathon

-Limitations of computing power to output desired outcomes

Accomplishments that we're proud of

-Developing our own model to detect guns across multiple streams of videos

-Using Google Cloud Platform to accurately obtain faces of possible suspects

-Using Amazon Web Services to send SMS notifications

What we learned

-Learned how to deploy ML models to a Web App

-Using Cloud Services for Machine Learning

What's next for No More Shootings #20

-To improve the accuracy of the gun detection

-differentiate between proper and improper usage of guns to reduce false alarms

Built With

Share this project:
×

Updates