Boosting intelligence with data

Executive Summary

Acoustic Threat Detection (ATD) has a longstanding history, and it continues to evolve. Our AI technology detects, locates and tracks gunshots in public places. The ATD sensors detect and localize a shot in two ways:

1. The first way is by the muzzle blast of a gunshot.
2. The second, and most important part of the technology, is the ability to detect the ballistic soundwave caused by the projectile in motion. Identifying the soundwave gives the sensor the ability to detect the exact location of the threat.

The artificial intelligence that powers our technology is optimized through learnings from the data collection that occurs when the sensors are in use. Each sensor adapts to its environment in order to understand what constitutes normal noise levels versus anomalies. ATD continues to learn as more threats are detected, so our team recently took a trip to our ballistic testing site to put it to use.

The Challenge

Since data collection improves the artificial intelligence that powers ATD, testing is paramount to improving our technology. Our team traveled to our ATD testing site in eastern Louisiana in August. Using four sensors and a variety of guns, we conducted tests for two days. Testing took place indoors and outdoors. Three of the four sensors were mounted 8 feet above ground level on smart city poles. The fourth sensor (sensor 1009) was mounted on a post outside of the testing center on the first day, but it was moved indoors on the second day. All of the sensors were wired to a PTZ camera that connected to a VMS system.

The Solution

We conducted testing by firing guns from different locations and in various directions. Sensors are capable of working independently of each other, so we collected information on each individual sensor. Though all of our sensors functioned properly during testing, we did encounter a few unique scenarios. ATD is wired to detect threats within a 500 foot radius. However, our technology is powerful enough that outliers can and do occur. We fired shots from approximately 600 feet out, and each of our sensors responded to the sound of the ballistic. Additionally, we tested to see how our sensors would respond when a shot was fired from inside of our testing facility. Sensor 1009, which was mounted inside during this test on the second day, responded promptly as expected. In addition, sensor 1008 also detected the indoor shot from outside. Lastly, the sensors were successful in detecting multiple shooters at once and shots fired at a higher elevation than the sensor itself.

While shots were fired, our team monitored the threat tracking information each sensor collected through our VMS. As each sensor localized the gunshots, the VMS showed us a view of the gunshot location from each camera’s perspective. We received text alerts featuring an image and description of the test shooter. These alerts are triggered by our VMS as soon as a threat is detected. As our test shooter continued to move through the testing site firing shots, the VMS continued to display the updated locations as detected by the sensors.

The Outcome

After two days of testing, our team returned to our headquarters in Richmond, VA. The data ATD collected while our team was in Louisiana will continue to improve its machine performance. Future visits to the test site will enable us to test additional abilities including ATD’s ability to detect shattering glass, fireworks, elevated voices, and more. This testing will support continued AI improvements beyond gunshot detection. As our ATD technology continues to evolve, we are committed to seeing that it never stops learning.