Gunshot detection has become a major request from many cities deploying smart city solutions. Current state of the art solutions requires a human interaction mechanism to detect the gunshot incident and number of shots through an operations center (OC). Based on this requirement, current solutions may introduce latencies ranging from 5-10 minutes from the moment a gunshot is detected to alerting of first responders.
According to the FBI, about 70% of active shooter situations are over in under five minutes, and the website of National Sheriffs’ Association states that “shaving even seconds off the notification and response times can result in vastly different outcomes in these situations.” The response time and latency introduced by existing solutions cannot be reduced as the OC in the loop introduces a human factor that needs to listen to the scene before calling and manually dispatching first responders.
In this paper we introduce a novel gunshot detection mechanism that is fully automated. Our paper describes the software platform along with algorithms that enable:
Our approach applies machine learning (ML) algorithms at the edge to detect and discriminate gunshots from indoor/outdoor ambient sound. When gunshots are detected, the sound direction is also derived, and security cameras follow the direction of the sound to capture the scene, identify the shooter(s) and automatically inform responders about the incident, thus providing significantly improved situational awareness.