From what I know augmented reality is a set of methods to generate information on heads-up displays (think "Robocop"-style displays). This field is currently divided into two major areas from an algorithmic standpoint:
- marker-based
- positional-based
Marker-based augmented reality is based on the computer processing artificial markers of in the real world (examples: QR codes, barcodes, or similar markers) and superimpose computer-generated images based on where the markers are located. This area requires significant image processing tasks done by the computer. Ideally the computer will recognize real-world objects directly and thus the markers are no longer needed, but it is still a topic that requires much research.
Positional-based augmented reality is based on where you are located, where you are pointing to (as in heading), and where are the objects of interest are located relative to you. The computer then will superimpose images on top of the real-world image gathered. The computer doesn't need to do much image processing (almost none at all) except for superimposing the generated image on top of the camera image.
Of course some applications are hybrids of these techniques.