Researchers Train Drones to Spot Violent Behavior

first_img UPS Wants to Bring Drone Deliveries to U.S. HospitalsWatch: Drone Captures Great White Sharks Interacting Off Cape Cod Stay on target Let us know what you like about Geek by taking our survey.center_img Big Brother is watching you—and he’s got some new toys.An international team of scientists are using machine learning and camera-equipped drones to identify violent behavior in public places.In a paper titled “Eye in the Sky,” a trio of researchers describe their real-time autonomous drone surveillance system (DSS), which relies on a $200 Parrot AR quadcopter and a mobile Internet connection.AdChoices广告An algorithm, trained using deep learning, can detect humans and estimate their posture, spotting those dubbed “violent”—which, in this case, includes strangling, punching, kicking, shooting, and stabbing.DSS can also monitor illegal border crossings, identify vandals, and recognize kidnappers.This system, scientists hope, will one day be scaled up to prevent crimes in open areas and crowded events.Lead study author Amarjot Singh, a research student at the University of Cambridge, was motivated by last year’s tragic Manchester Arena bombing, he told The Verge.Similar attacks could be prevented in the future, he said, if surveillance cameras can automatically recognize suspicious behavior.As The Verge pointed out, though, Singh et al. are far from ready to deploy their drone system.The team reported 94 percent accuracy with only one violent individual per image. But that number begins to slip as more people enter the frame: It falls to 84 percent when looking at five individuals, and 79 percent for 10.The violent individual detected by the DSS framework is shown in red, while the neutral human is shown in cyan (via Amarjot Singh et al.)Those numbers, however—based on volunteers pretending to attack each other—don’t necessarily represent the real world or the crimes that occur in it. Which is why researchers plan to test their “eyes in the sky” during two upcoming festivals at the National Institute of Technology, Warangal, in India.“We have permission to fly over [Technozion], happening in a month,” he said. “And are seeking permission for [Spring Spree].”It’s unclear how the DSS will perform in the wild, when video footage is blurrier, crowds are bigger, and actions are more subtle; the paper does not provide data on false-positive rates.Imagine the autonomous drones mistaking a high-five among friends for a dangerous right hook. Or accidentally sicking the police on a group of senior citizens practicing tai chi in the park.Read the full paper—co-authored by Singh, Devendra Patil of NIT Warangal, and SN Omkar from the Indian Institute of Science—online.last_img read more