Automated surveillance drones will become more common as researchers and companies discover new ways to use machine learning to analyze live video footage. A recent project from experts in India and the UK show one possible use for this new technology, which is identifying violent people in crowds using drones equipped with cameras and custom software.

Researchers explained more about the system in a paper called “Eye in the Sky”. It makes use of a simple Parrot drone. The drone only costs about $200 and sends video footage over a mobile internet network for real time processing. They then use an advanced algorithm to analyze the postures and behaviors of people in crowds, looking for people who are in “violent” poses.  The software analyzes a crowd of people looking for people in specific poses such as punching, kicking, and stabbing to name a few.

Researchers are hopeful that this system will be used to detect violent people in public areas or at large events. Amarjot Singh is the lead researcher for this project at the University of Cambridge. In an interview, he stated that he was motivated by events like the 2017 Manchester Arena bombing. Singh stated that these types of attacks may soon be prevented if surveillance drones can spot unusual behavior such as leaving a bag unattended. The AI software can analyze a crowd and notice if someone puts down a bag and walks away from it thereby alerting authorities.

The research team stated that the system was not 100% accurate at recognizing violent behaviors. However, they also explained that the bigger the crowd, the lower the percentage as there is more people to analyze. More importantly, these figures do not fully represent real world applications. To accurately train the AI drones, the researchers used video clips of volunteers acting violently in a group of people while the software recorded and analyzed them.

Although this was not a perfect example of how the surveillance system will work in the real world, the researchers plan to test the drones during some upcoming festivals in India.  In a real life situation, video footage can be blurry, the crowd size can also be larger, and people’s actions can be misinterpreted.  The more real world experience that the AI software analyzes, the better it will become and hopefully preventing crimes and violence before they occur. The real world application for this software has lots of benefits and could become common around the world whenever people gather in crowds.