AlertNow: A Real-Time, Low Cost, Multi Modal Fusion System to Detect Gun Threats in US Schools

Gun violence is a dangerous epidemic in the United States. Every year, more than 45,000 people die from guns, and the problem is only growing worse, with a 34% increase in gun related violence in the last decade. Technology can play a pivotal role in combating the gun violence crisis in the US schools by deploying advanced computer vision and deep learning models to detect firearms and gunshots in live video surveillance systems. Published literature demonstrates that the current commercially available systems specialize in either gunshot detection using acoustic signal or firearm detection using visual data only. This research hypothesizes that an integrated system combining both acoustic and visual modalities can enhance the accuracy and robustness of a gun detection system. The purpose of this research is to develop a cost effective, reliable, and multi-modal fusion-based prototype solution, to detect gun threats and notify authorities in real-time. AlertNow is developed by integrating CNN model for gunshot detection and YOLOv8 model to detect firearms. This multi-modal prototype system is trained, validated, and tested using extensive datasets to enhance detection accuracy with an ability to trigger real-time alerts for timely intervention by security staff in critical situations. The results indicate that AlertNow system estimated performance is superior to other approaches used to gun detect threats with an overall precision, recall and F1 score greater than 99%. The integrated system also displayed a remarkable low latency across all notification delivery channels and therefore making it a viable solution for real-time alerts.