Czech scientists develop ‘drones with brains’ to make crowd surveillance far easier

Fights and injuries occurring inside large groups of people could soon be easier to detect, thanks to a new Czech-developed crowd surveillance system that analyses drone footage via the use of neural networks.

The innovative new system was developed through a joint research programme between the Brno University of Technology and the Czech Police, who have increasingly turned to drones to monitor large crowds in recent years.

Existing drones, however, are nowhere near as efficient as they could be, according to David Bažout, a fresh graduate from the university’s Information Technology faculty.

“Police drones that are currently used above areas of interest do not have any training data. No one is able to assess what is normal behaviour and what not.”

Bažout therefore decided to develop a neural network system that analyses drone footage in a way similar to the human brain.

Video footage recorded by the drone is divided into several smaller “cells” that are then analysed by the network which establishes a general picture of what is going on. A model of standard behaviour in the given environment is then developed, and any anomalies to the pattern are reported to the observer by highlighting the area in red colour.

The advantage of the system developed by Bažout is that it can learn in real time, without receiving any learning data before being sent out into the area of interest. He points to one of the tests of the system, taken by scientists at the Brno University of Technology, as an example.

“This group of children kept running counter-clockwise. Then they suddenly stopped and started running in the other direction. The system marked this change in behaviour with a red circle, so that any person in charge of monitoring could take note and react.”

In another test, the system had to observe players on a football pitch. Suddenly, some of the players had to lie on the ground. The anomaly was again marked in red by the neural network.

There are of course many instances of unusual behaviour in large crowds. To avoid the screen constantly turning red, Bažout and his team added an option allowing the operator to set the sensitivity of the system to detected changes.

The Czech Police are now looking into starting their own tests in real crowd control situations. The ultimate aim of using this improvement in video analysis is to enable officers or ambulance services to reach dangerous areas faster. The system could also make it harder for criminals to blend into large crowds, for example by highlighting sudden accelerations in movement.

David Bažout says that the system also has the potential to be expanded into areas beyond crowd control, such as road crossings.

Authors: Tom McEnchroe , Barbora Kroutilíková
run audio