SLAM (simultaneous localization and mapping)

Tracking a moving camera's own pose while building a map of the scene, from a live stream, at the same time.

SLAM (simultaneous localization and mapping) tracks a moving sensor's own pose and builds a map of its environment concurrently from a live stream, rather than SfM's offline batch of images. Because small per-frame pose errors accumulate into drift over a long trajectory, a practical SLAM system relies on loop closure -- recognizing a previously visited place and using that recognition to correct the accumulated drift across the whole map.

Examples

A robot vacuum's SLAM system builds a map of a new room while simultaneously tracking its own position within that same map, correcting drift whenever it recognizes a spot it has already passed.