Keypoint Matching

Keypoints (aka features) are things (patches of pixels, really) that are visible on more than one image. Finding identical keypoints and tracking them across a series of images allows Structure from Motion software to extract 3D information from the images.

In classical photogrammetry, people would identify just a handful of such keypoints across a rather small series of images (sometimes just two), but with very high, manual accuracy. Automated computer vision approaches extract thousands of keypoints even from low resolution images. Each individual keypoint may not be tracked with great accuracy, but this is compensated by using massive numbers of them. So much compensated in fact, that things like lens calibration, which is an important part of classic photogrammetric techniques, must no longer be done manually.