Basically, they use an infrared projector in order to project (the surprise!) a number of infrared dots (usually 6 of them) on some object, serving as a marker. The hand held device is composed of two cameras, one scenic, and another infrared one. The IR camera is looking for the projected dots, and calculates the distances between them. The device calculates using those distances the alignment and position of the augmented object, and thus can annotate images taken by the scenic camera. Which is pretty cool.
The obvious downsides of this method is the use of two cameras (though, I don't think it's impossible to combine them both into one), and the limited number of markers. Unlike the visual makers we all learned to hate, with their many white and black bits, with the IR markers, one cannot simply rearrange the "dots" in order to get a new marker, since the dots' positions determine the direction of the object. Maybe adding few more dots to the jumble can relieve this problem. Another major problem with this technique is that a projector can only mark a limited number of objects, so if we wanted to implement this method in order to augment museum items, we would need numerous projectors.
Wang, Liu, Wang, Infrared Marker Based Augmented Reality System for Equipment Maintenance, International Conference on Computer Science and Software Engineering, 2008