With the increasing outdoor crime rate in recent years and the recent surge in support for more objective law enforcement, our team decided to create an automated solution that detects crimes. Additionally, automation of crime detection could potentially eliminate bias, which is another major problem in our society.

What it does

SkySecure uses Machine Learning technology to detect crime in photos taken by a surveillance drone. The SkySecure app displays the location of these crimes on a map.

How we built it

We used a camera dji tello drone that one of our members had and connected the camera to a database. We then wrote machine learning code that would take the photos taken by the drone and detect possible crime or suspicious activity in them. Then the locations that qualified as possible crime locations would be sent to a Firebase database to be stored. The Firebase database is connected to our android app. The android app reads those coordinates from Firebase and converts them into locations on a map generated by Mapbox. The machine learning algorithm is a lightweight keras multi-layered-perceptron classifier that uses imagenet yolov3 as a feature extractor. Because the model is lightweight, it can be easily trained on verified data from the app. So, the more you use the app, the more accurate it becomes.

Challenges we ran into

Configuring Mapbox with our android app proved to be a challenge because the current documentation was not up to par and it wasn't accomplishing exactly what we had wanted. Because of time constraints, we weren't able to create a good image crime dataset (we couldn't find a good one online).

Accomplishments that we're proud of

We are proud of the fact that we were able to connect all these systems together and provide our user with an easy to use app to monitor this process.

What we learned

We learned how to integrate a robust machine learning pipeline running on a dji tello drone with an android studio app.

What's next for SkySecure

We were held back by our hardware capabilities a little bit so we would definitely want to upgrade our equipment to provide precise GPS locations and crisp camera shots.


YES WE KNOW THAT HOLDING A KNIFE ISN'T ILLEGAL. Since we couldn't find a dataset on images of crimes being committed, we had to create our own. None of us exactly had the means to simulate illegal violent crimes, so this was the closest we could get.

Share this project: