Shotspotter Patent Applies Machine Learning Accuracy for Gunshot Detection
Shotspotter acoustic sensor.
ShotSpotter reports the U.S. Patent and Trademark Office (USPTO) has granted it U.S. Patent No. 10,424,048 entitled “Systems and Methods Involving Creation and/or Utilization of Image Mosaics in Classification of Acoustic Events.”
ShotSpotter’s real-time gunshot detection solution uses a 2-step process that employs machine classification and human review. The system can distinguish with high accuracy whether a loud, impulsive sound detected by its acoustic sensors is a gunshot or a non-gunshot incident, such as fireworks, in less than 60 seconds, according to the company.
The innovation behind the patent granted to ShotSpotter covers the conversion of multiple features of the audio event into a set of visual displays that are combined into a single image mosaic. This enables the system to leverage deep learning neural networks that typically identify and classify images, not sounds.
Since its implementation, this approach, along with the ability to train the system using ShotSpotter’s proprietary database of over 14 million gunshot and non-gunshot incidents, has resulted in significant improvements in machine classification accuracy, Shotspotter said.
Events that are machine-classified with high confidence as non-gunshots can now be excluded from the human classification step. This allows more focused expert review of probable gunshots in ShotSpotter’s Incident Review Center (IRC) prior to notifying patrol officers digitally about the gunfire incident location. Additionally, the company expects to benefit as its customer base grows by being more efficient in the staffing of the IRC.