Visual Odometry Using Multiple Landmarks

Motivation

The experimental validation of different control strategies as well of computer vision algorithms aimed to solve specific problems of mobile robot navigation needs, in general, to have a reliable measure of the robot pose (location and orientation) respect to some predefined reference frame. Here, robustness must be required in order to deal with the usual problems of variable illumination conditions and/or significant view-point changes.

General (brief) description

To solve the above mentioned problems we propose to integrate the multiple measures of the robot pose relative to a set of artificial landmarks distributed over the environment on which the robot moves. Such landmarks, also called "fiducial markers" in the literature, corresponds to simple black and white planar patterns that are easily detectable in real-time by standard image processing methods. Once their are detected, the knowledge of the camera parameters (obtained off-line) and the pattern physical dimensions will allow to recover the pose of the robot relative to each one of such landmarks. This individual measures, together with their uncertainty estimates, will be used to compute the instantaneous pose of the robot in a robust way, e.g., using a non-linear Kalman filter.