Man (probably James Bond) holding a gun

Machine learning system detects guns in real time

Image credit: Dreamstime

Researchers from the University of Granada, Spain, have developed a smart computer system to detect when a gun is drawn in real time using video footage.

The system uses a deep neural network: a complex artificial simulation of the human brain which ‘learns’ to identify patterns by processing huge amounts of data, either with or without human supervision. This can allow for the rapid analysis of language, scientific images, video and other, subtle forms of information, such as facial expressions.

The University of Granada researchers decided to train their software to recognise the drawing of a firearm, using a dataset of low-quality videos from YouTube, and cult 90s films featuring trigger-happy characters, including Pulp Fiction, Mission Impossible and James Bond films.

Using this data, they were able to train their system to detect guns in video with over 96.5 per cent effectiveness. The programme can analyse five frames a second, rendering it capable of real-time gun detection.

When a gun appears in the video, the programme sends an alert, which appears as a red box surrounding the weapon on the screen.

Professor Francisco Herrera of the University of Granada suggests that this could be combined with video cameras and an alarm system for unmanned supervision of a sensitive area.

While metal detectors have the advantage of being able to detect a hidden weapon, they are, in many ways, an inefficient means of preventing danger. A gun can only be detected if it is carried through the gates, and if it is made of metal; the rise of 3D printing means that this is not always the case. Preventing an attack also requires constant supervision by a human operator.

In the future, a computer system for detecting guns in real time in video feeds could complement current security technology like metal detectors in airports and other public places.

The system could also prove helpful in detecting violent video content on the internet. Following several incidents of murders, suicides and other attacks being shared on social media in recent months, Facebook and other tech giants have come under increased political pressure to tighten restrictions on violent or otherwise abusive content uploaded to their platforms.

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them

Close