Applying Machine Learning to Early Weapon Detection

​

​

James Helm, Alex Green, Nathan Cheney

Rising Gun Related Deaths

​

• Gun related deaths have been on the rise in recent years

​

• Exceptional rise during the pandemic and recent social strife incidents

​

• Can Machine Learning be utilized to support an early warning system?

Image: https://www.thetrace.org/2021/12/gun-violence-data-stats-2021/

Police Response Time

​

• Average Police response time to school shootings is 18 minutes*

​

• 18 minutes can be an eternity, early threat detection can accelerate the response

​

Image: https://www.erienewsnow.com/story/30262486/local-police-take-part-in-active-shooter-drill-at-former-erie-school

*Study: https://www.creditdonkey.com/average-police-response-time.html

Leveraging Computer Vision

​

​

• Machine Learning Computer Vision models can provide an extra set of eyes

​

• Easily adoptable to existing Home/Public/Work camera security systems

Using Pre-trained Models

​

​

• Using YOLOv5 as a base pre-trained model we were able to create a model that can detect pistols and knives in a relatively quick time period

​

• Using pre-trained models and data allowed us to quickly develop the application

Source: https://pytorch.org/hub/ultralytics_yolov5/

Obscured Images

​

​

• The model can identify partially obscured weapons

​

• Accuracy drops with higher obscurity/masking

Multiple Object Detection

​

• Model is currently designed to identify both knives and pistols

​

• Can be easily adapted to account for other weapon types (E.g.; rifles, rocket launchers, etc.)

False Positives

• No Machine Learning model has 100% accuracy which leads to possibilities of false positives creating un-wanted alerts

​

• Extensive model training mitigates this, but it will always be an issue

​

• Secondary human verification can serve as the best of both worlds

Intentional Masking

​

• What happens when a gun doesn’t look like a gun?

​

• Understanding how weapons are detected opens the door to hiding weapons unconventionally

​

• If we become reliant on automatic detection, we may miss the true threat

Source: https://www.americanrifleman.org/content/6-unique-handguns-that-break-the-mold/

Source: https://www.engadget.com/covid-19-youtube-diy-mask-launcher-154046116.html

Questions?