DetectionDetecting threats in a crowd

Around a military camp situated close to a built-up area there are always people moving about. Scientists at FOI, the Swedish Defense Research Agency, have therefore created a multi-sensor system designed to be able to detect threats by identifying unusual patterns of movement involving individuals or groups.

The technology is based on a number of sensors that detect human movement patterns which are then interpreted in a signal processing system. When the patterns deviate from what might be expected, the system triggers an alarm.

For example, it might be that one or more individuals in a crowd are moving more quickly, or in a different direction, compared to the general flow, which could mean that something irregular is going on. Or perhaps a group is loitering by the side of a road, which could indicate that a roadside bomb is being planted.

A FOI release reports that an important part of this research is about optimizing combinations of sensors and their positioning.”

“All sensors have their advantages and disadvantages. A visual sensor is fine during daylight but not at night. An infrared sensor is good when it is dark but not when it is too hot. An acoustic sensor captures all sounds and cannot be blocked out, but different sounds can be difficult to separate,” says Maria Andersson, senior research scientist at FOI.

She adds: “If one knows what one is looking for, then it is not too difficult, a fight can be converted into known algorithms. It is more difficult when the system has to interpret what a gathering of people means, or the way in which groups form or disperse.”

Andersson tells how the system is now being developed to enable it to find groups who are on the move and to recognize an individual person’s movement patterns and how these relate to one another. The aim is to produce a system demonstrator during the course of 2014, the test results of which can then be commercialized by partners in industry.

“In the future we would like to develop the detection element to enable it to follow individual persons in a crowd. This is difficult, not least because people block each other’s sightlines. What we have to do, therefore, is to create methods of finding stable distinctive features,” Andersson says.

The closer we get to personal identification systems, the more important questions of data integrity become.

“We are working on this particular aspect in various EU projects. Provided that detection is taking place in order to tell whether something unusual is taking place, the question of personal data is not so sensitive. If individual persons are being tracked we can save the information in the system for a while in case something happens but we must then delete it. We are also looking at the possibility of finding forms of detection that do not identify human features but rather focus on the color of clothing or some other characteristic,” says Andersson.