Localisation methodology by image processing for mobile robot navigation in outdoor environment

The work described in this paper deals with the use of the artificial vision for localising a mobile robot in an outdoor environment. In particular image acquisition technique, based on a two-axis camera platform, a CCD camera and its controlled zoom lens, is utilised. At a first step, an image is caught by means of the zoom lens tuned with a short focal length. This enables the detection of the image main features. Then, at a second step, another image corresponding to a long focal length is caught from the same place, in order to improve the localisation accuracy of the image features. The selected robot localisation methodology consists in using two different types of vision representation. The first one is the result of a data processing reduced to its most vertical rectilinear indicators applied on the images pair corresponding to the both focal lengths. The second representation is based on a prediction constructed from a memorised digital map. The robot localisation is obtained firstly by matching the two previous images and then by a triangulation calculation. Finally, some experimental results are presented, followed by a discussion about the quality and the efficiency of the method.<
>