Abstract

Ansel was developed to help protect public safety utilizing existing security cameras and infrastructure for firearm detection. The system utilized YOLOv5, an open-source real- time detection algorithm that was trained on over 3,000 images to identify firearms with boundary boxes and alerts on CCTV footage. I trained the model on a custom dataset and weights, labeling every image for over 30 hours. The system was integrated with a real-time detection map for viewing on a website. The website displayed real-time alerts indicating threats where the cameras could see in proximity to users. My project prioritized privacy by keeping all data on one server in a closed loop, only needing to go outside the network once if there is an alert that a firearm was seen. Ansel’s predictive capabilities show a proactive approach for early detection and have been able to demonstrate early detection and for victims and first responders.