Drones Using AI to Spot Violent Behavior in Crowds

Researchers are pairing artificial intelligence and drones in the latest experiment to prevent crimes before they happen.

• By Sydny Shepard
• Jun 11, 2018

Researchers are pairing artificial intelligence and drones in the latest experiment to prevent crimes before they happen.

In a paper titled, "Eye in the Sky" the researchers describe a system that uses a drone to transmit video footage over a mobile internet connection for real-time analysis. An algorithm trained using deep learning estimates the poses of humans in the video and matches them to postures the researches have designated as "violent."

For the purposes of the project, just five poses are included: strangling, punching, kicking, shooting and stabbing. To train and test their AI, the researchers had their own video clips of volunteers pretending to attack one another. The volunteers are generously spaced apart and attack one another with exaggerated movements.

The researchers have acknowledged that this isn't a perfect system and as of now their research was only 94 percent accurate in identifying violent poses. (The more people in the frame, the system was only79 percent accurate in finding the violent poses.)

The next step is to use the system in real-life events. There can often be a large disconnect between stages and real events, but researchers hope to test and teach the AI to learn more about the video being taken at these events. 

The researchers hope that the system will be used to detect crime in public spaces at large events. Lead researcher Amarjot Singh told The Verge that he was inspired to create the system after events like the Manchester Arena bombing in 2017. Singh told reporters that attacks like this could be prevented in the future if surveillance cameras can automatically spot suspicious behavior, like someone leaving a bag unattended for a long period of time.