SafeTogether

A lens into a world of preventing violence.

Comment 1

Inspiration

On average, 24 people per minute are victims of rape, physical violence, or stalking by a partner in the United States, 12 million women and men over the course of a year. In the last two months, these cases have increased significantly. In California, for example, domestic abuse calls to the Fresno sheriff’s office increased by more than 70%; in San Antonio, Texas, there was a 21% increase in family violence calls, with more than 500 additional calls during the first three weeks in March compared to the same period last year. Our zeal to make the world a better place for everyone motivated us to work on this problem and come up with potential solutions.

What it does

SafeTogether detects two particular kind of hand gestures that together is known as a signal for help. It detects and alerts local help or the authorities about that person being in danger.

How we built it

It is built using Auto ML object detection model from google vision API. We collected images of both signs from google and labelled them as either “four-fingers-up” or “thumb-in-fist” using Google’s Auto ML object detection model. Precision of the model was 100% while Recall was 85.71%. The training data consisted of only 25 images but still the model worked pretty well.

Challenges we ran into

  1. Small training data : Due to less time, we couldn't work on a larger dataset.
  2. There was no such dataset available for "signal for help". So, we manually collected images and labelled them.

Accomplishments that we're proud of

We are proud to have met our talented and outstanding teammates. As a team of mostly first time attendees we were able to identify a project to work on and develop something in only a 24 hour time span on our own with the help of some wonderful HackViolet Mentors.

What we learned

We learned how to set up, train, and use Google's AutoML Vision, as well as learned to use and better navigate the Google Cloud Console.

What's next for Test

  1. We will work on larger dataset.
  2. Test this real time model with different apps like youtube.

Built With

Share this project:

Updates