-
-
Pistol with 73% Detection Confidence
-
Mean Average Prediction (mAP) Value of Warden
-
Pistol with 86% Detection Confidence
-
Multiple Pistols Detected by Warden
-
Pistol with 79% Detection Confidence
-
Graphs Depicted Warden's Performance
-
Multiple Oistols Ddetected By Warden
-
Example Images Needed To Be Annotated Then Used To Train
Inspiration
Every day, over 300 Americans are affected by gun violence, whether it be as a witness or a victim. With the ongoing school shootings and mall shootings, the need to find a solution to stop gun violence is greater than ever. Fueled by inspiration to change this, our team developed Warden.
What it does
Warden is an artificial intelligence model trained to detect pistols in real time (through a working webcam or camera). If there is a pistol in the sight of the camera, Warden will highlight a box around it and also mark it as a pistol. Warden is also well trained in detecting weaponry that is not a pistol.
How we built it
Warden was built using Roboflow's pistol image data set as well as other artificial intelligence features provided by the site. Our website was built using HTML and CSS. We first spent 3 hours manually annotating around 3000 images with pistols in them in order to train our model. Then, using the trained set, we tested the model through our webcam and using online images.
Challenges we ran into
One challenge we ran into was deciding whether or not to include front-view pictures of pistols in our annotations, as they can be misinterpreted for other guns.
Accomplishments that we're proud of
We are proud to have achieved a Mean Average Prediction score of over 85% and a recall rate of over 80%. Furthermore, we are proud to have the model detect multiple pistols in a single image which can prove helpful in various situations.
What we learned
Our team learned that we can impact a lot of people from our computers and if we have the will power and drive, we can innovative solutions that can impact on a broader scale. We also learned the importance of communication as well as what each team member's strengths and weaknesses were.
What's next for Warden
Warden is still in its development phase. We hope to refine our model even more by using more images to improve our prediction and recall scores. In addition, we hope to broaden Warden's scale to not only include pistols but other harmful weaponry.
Log in or sign up for Devpost to join the conversation.