On-line estimation of the reference visual features application to a vision based long range navigation task

In this paper, we consider the problem of realizing a vision-based navigation task. To this aim, we present an algorithm allowing to automatically compute the necessary reference visual features. This algorithm relies on a predictor/estimator pair able to determine the visual features depth sufficiently rapidly with respect to the control law sampling period. The proposed method can then be used on-line. An example of such an application is presented through the realization of a vision-based long range navigation task. Both simulation and experimentation results validate the whole approach.