One-track mind

BAE Systems has developed a CCTV camera system that it claims can autonomously track individuals even if they change their appearance or try to hide in a crowd.

The technology is a result of research carried out by the defence giant as part of a 10-partner EC project on integrated surveillance in crowded areas.

By analysing TV images gathered from a single, standard pan/tilt/zoom camera, the software produces templates, squares of four by four pixels of the image which appear as coloured dots scattered around a target’s body. It then tracks a person by following the position of these templates as he or she moves through a scene.

‘The identification of individuals is done manually. The user can see the image from the camera and, with a mouse, click on somebody to say “follow this person for me automatically” from the click onwards,’ said project manager Andrew Cooke.

‘We then have a step which generates all of these templates which we use to track the target. Normally there are 25 of these very small patches which are dotted around a person, so some of them may be on the head, some on the torso and some on the legs.’

Cooke said that the system adapted to changes by ‘throwing away’ templates that no longer apply, such as when they are too far away from the centre of the target, and introducing new ones.

‘The beauty of the system is that it does not rely on static templates of people. You can imagine the problem of individuals moving through crowds is that their appearance can change when taking off their jacket or when the light changes. So we have tried to incorporate an adaptive element to the tracking. This means that if a jacket is take off we can quickly adapt the tracking to accommodate that fact.

‘The coloured templates characterise the target, so light green patches are the ones we are using to track the target, dark blue would represent the templates we would ‘kill’ or throw away, and red patches would be the new ones we would introduce,’ he said.

Another feature of the software is its ability to detect suspicious behaviour.

‘What we have essentially done is produce a system which can alert operators to some loitering behaviour. So if somebody is hanging around in an area declared as sensitive or high security, there is an underlying process which can provide alarms for that sort of situation,’ said Cooke.

BAE Systems developed two versions of the system. The first is real-time technology that, at present, can only track one person at a time using the pan/tilt/zoom camera. The second version uses the same processing, but in a slow-time mode that can track multiple targets for post-event analysis. Cooke said that a future development of the real-time version could enable another camera to pick up the trail from the initial tracking camera.

Now awaiting certification, the technology was successfully demonstrated at Elsag, an Italian-based systems integrator, by showing that it could track someone joining and then leaving a group.

‘The crowded environment is obviously the most challenging aspect of tracking — you have many occlusions, lots of people looking very similar. There are many automatic tracking systems that exist today but in my opinion none of them addresses the crowded areas situation,’ said Cooke.

‘Some of them track by taking differenced images and extracting the movement by subtracting images from each other. So what you end up with is a binary image, and you can use the motion of that image to track people. But that technique becomes extremely hard when you approach high-density environments.

‘Other systems use a fixed template approach, which means that if someone takes their jacket off, for example, they become a completely different person from an algorithm perspective.’

The EC project, integrated surveillance of crowded areas for public security (ISCAPS) started in 2005. Its aim was to analyse the climate of international security and develop technologies to increase the quality of existing methods of identifying and tracking people, goods and vehicles.

Paris-based Sagem, for example, concentrated on developing biometric reading technology, while Reading University developed a large multi-camera tracking system.