Vision based Robot Localization using Invariant Features

In order for a mobile robot to determine its current position using visual information from the surrounding environment, it must maintain a representation of a visual map and be able to keep track of it. Such a problem, usually referred to as robot localization, is worked out here using different invariant features extracted from images that are captured by the robot during its navigation.

The technique works like this: The robot, assisted with a camera, establishes a database of features that is maintained during its navigation, each feature corresponds to a position or landmark.  When losing its location, the robot will extract a new feature vector from the current position and compare it with those in the database. Once the most similar vector is found, the robot can retrieve its position.

Matching local invariant features between two images.