System for automatic monitoring increasingly penetrate into our lives, as developers discover new ways to use machine learning to analyze the video stream. A new project of researchers from the UK and India allows the use of this technology for the determination of aggressive behavior in crowds using feature cameras drones.
The developers used a simple Parrot quadcopter for $ 200 for video transmission over the mobile channel for further analysis in real-time. The algorithm using the machine learning trained to identify “aggressive” posture people.
At the moment, for the purposes of the project are choking, punching, kicking, shooting and stabbing. For a small number of participants, the accuracy reaches 94%, but when the number of participants grows to ten, the accuracy drops to 79%.
According to the project developers, their system can be used to identify crimes in public places and during mass events.