Acquisition and Recognition of Natural Landmarks
for Vision-Based
Autonomous Robot Navigation

Salvatore Livatino

Computer Vision and Media Technology Laboratory
Aalborg University, Denmark
Email: sl@cvmt.dk


Abstract

The use of landmarks for robot navigation is a popular alternative to having a geometrical model of the environment through which to navigate and monitor self-localization. If the landmarks are defined as special visual structures already in the environment then we have the possibility of fully autonomous navigation and self-localization using automatically selected landmarks.
The developed PhD project investigates autonomous robot navigation and proposes a new method which benefits from the potential of the visual sensor to provide accuracy and reliability to the navigation process while relying on naturally available environment features (natural landmarks). The goal is also to integrate techniques and algorithms (also related to other research field) in the same navigation system, in order to improve localization performance and system autonomy.
The proposed localization strategy is based on a continuous update of the estimated robot position while the robot is moving. In order to make the system autonomous, both acquisition and observation of landmarks have to be carried out automatically. It is consequently proposed a method for learning and navigation of a working environment and to explore automatic acquisition and recognition of visual landmarks. In particular, a two-phase procedure is proposed: first phase is for an automatic acquisition of visual-landmarks, second phase is for estimating robot position during navigation (based on the acquired landmarks).
The feasibility and applicability of the proposed method is based on a system with a simple setup. The novelty and potentiality are in combining algorithms for panoramic view-synthesis, attention selection, stereo reconstruction, triangulation, optimal triplet selection, and image-based rendering. Experiments demonstrate that the system can automatically learn and store visual landmarks, and later recognize these landmarks from arbitrary positions and thus estimate robot position and heading.