New Tech Automatically Spots Guns in Videos

The algorithm works in real time and can greatly enhance the capabilities of video security systems, but it could also be used to monitor guns on social media.
The algorithm works in real time and can greatly enhance the capabilities of video security systems, but it could also be used to monitor guns on social media.photo from phys.org

A new technological breakthrough may change the way security systems, and content censors on social media, identify firearms in videos—and the developers used images of guns in movies to create it.

According to this story from phys.org, scientists at the University of Grenada have designed an artificial intelligence-based computer system that automatically detects when a subject in a video draws a gun, in real time.

Obviously, this could have a huge impact on the security industry and its capability to monitoring large public areas like airports and malls while potentially cutting down on response time. In some cases, it could even raise alarms before a bad actor has a chance to hurt someone, as it can be integrated into existing security and alarm systems anywhere cameras are in place and doesn't require any supervision, the story says.

However, the story also says the content can be used to monitor guns on videos uploaded to social networks like Facebook, Youtube, and Twitter. And since it doesn't require a human being to actually operate, a high-powered version could conceivably automatically flag any video with a handgun, regardless of its nature. Similar technology has already been implemented online to filter out pornographic or violent material, though most require real-live people to make the final call.

The tech comes from researchers Francisco Herrera Triguero, Roberto Olmos, and Siham Tabik, who work in the school’s Department of Computational and Artificial Itelligence Sciences.

The story says the team analyzed low-quality videos from YouTube and movies from the 1990s, like Pulp Fiction, Mission Impossible, and a number of James Bond films to develop their algorithm, which has a 96.5 percent effectiveness rate and is capable of analyzing five frames of video per second in real time.

From the information in the story, it seems the algorithm has only been training to recognize handguns in videos…so far.