This analysis creates a unique fingerprint of the point and it's surroundings within the image. If the feature's print is found in another image it can then be matched, providing a basis for calculating the relative positions and rotations of features in a physical environment.

Incremental errors along the device’s movement path make precision impossible. Drift gradually comes about where the device pose and cloud points start appearing out-of-place. To further complicate, drift errors escalate when devices heat-up as their sensors are calibrated to work optimally in a specific temperature range. .

A loop closure occurs when a point cloud region is revisited. When a closure takes place it mitigates the drift problem as the matched points in the intersected region correct their drifted poses. Other cloud points also shift accordingly in a chain reaction, minimising the accumulated errors and correcting the shape of the cloud to closer match the environment it resembles.

The cloud is not only used for tracking but relocating too. After losing tracking (lens covered, excessive motion between frames, low lighting) a device can near-instantly figure out where it is. It reorients itself, snapping to position by matching feature points from the current frame with features cached in the point cloud.