Move over Kinect, there's a new bit of body-tracking kit in town.
Humans learn more about another's appearance the more we look at them, and store this information away so we can recognise them the next time we see them. Now a smart camera has been developed that can do the same thing, allowing it to track individuals as they move in and out of video footage, or recognise their face or hand gestures.
Facial recognition systems can identify a person when they are looking directly at the camera, but tracking people as they move in and out of frame remains a difficult task.
The Predator camera constantly collects details of the person or object it is filming, allowing it to build up a model of the target, says its developer Zdenek Kalal at the University of Surrey in Guildford, UK.
The first time the system "sees" a person or object, it creates a model of it, which it records in its memory. Then, as it continues filming, it adds fresh details of the object from slightly different angles, building up a three-dimensional representation. This allows it to recognise the object again even if it leaves the shot and then reappears at a different angle.
As well as allowing police and security forces to track individuals through CCTV footage, the system could also help disabled people to control computing devices through facial expressions or gestures, says Kalal. Since the system learns as it goes, it would not need to be laboriously trained to work with one individual or gesture, but could adapt to each person's preferred method of control.
The system operates with just a simple webcam, and so could also be used for low-cost motion capture. Unlike Microsoft's motion-tracking system Kinect, which is only able to follow body parts at a certain distance from its camera, the Predator can track any object within its shot. The system could also be used in collision-avoidance systems for cars, to track other moving objects or vehicles.
Helen Knight, technology reporter
No comments:
Post a Comment