A Closer Look
Accurately detecting and precisely locating gunshots is a challenging problem given the many obstacles of buildings, trees, echoes, wind, rain and street noise. ShotSpotter has met the challenge with our innovative technology, developed over 20 years and protected by more than 30 patents for intellectual property. Our patents can be found here.
DETECTION & LOCATION
First, highly-specialized software analyzes audio signals for potential gunshots.
- The software filters out ambient background noise, such as traffic or wind, and listens for impulsive sounds characteristic of gunfire; we call these pulses.
- If the sensor detects a pulse, it extracts pulse features from the waveform, such as sharpness, strength, duration and decay time.
- If at least three sensors detect a pulse that is believed to be a gunshot, the sensor then sends a small data packet to cloud servers where multilateration is used based on time difference of arrival and angle of arrival of the sound to determine a precise location.
After the software determines the location of the sound source, it analyzes the pulse features to determine if the sound is likely to be gunfire.
- To evaluate and classify the sound, algorithms consider the distance from the sound source, pattern matching and other heuristic methods.
- The machine classifier compares the sound to the large database of known gunfire and other impulsive community sounds to determine if it is gunfire.
- Once an incident is classified as likely gunfire, it is sent to acoustic experts in our Incident Review Center (IRC) for additional analysis and it is published to police or dismissed.
The system pushes the alert notification to law enforcement and emergency responders.
- Incident notifications are triggered when the incident is confirmed as gunfire.
- Gunfire alerts are pushed to ShotSpotter mobile apps, and desktop and browser apps.
- The entire transaction from initial gunfire to alert takes place in less than 60 seconds.
Artificial Intelligence and Machine Learning
Investments in algorithms, artificial intelligence, machine learning and more recently deep learning have enabled us to continuously improve our ability to accurately classify gunshots. With more than 15 million incidents reviewed to date, ShotSpotter is in a unique position to feed this trove of data into our system for increasingly smarter and more precise results.