This work focuses on the navigation of a mobile robot equipped with a camera in a cluttered environment. To perform such a task, we propose to use the image based visual servoing (IBVS). This method consists in designing a control law using visual features provided by the camera. These features are defined by the projection of a characteristic landmark on the image plane. The IBVS based navigation requires to address three issues : the robot security with respect to the obstacles, the management of the occlusions and the long range navigation realization. Our contributions are mainly focused on the two last mentioned problems. First, we have dealt with the visual features estimation problem during occlusions. As the visual features depth is an important parameter in this process, we have developed a predictor/corrector pair able to estimate its value on-line. This method has provided nice results, even when the used measures are noisy. Second, we have considered the problem of performing a long range navigation with an IBVS. However, classically, using this kind of controller greatly limits the realizable displacement because the reference landmark must be seen from the beginning to the end of the mission. To relax this constraint, we have developed a topological map and a supervision algorithm which have then been coupled. The first one contains the most characteristic landmarks of the environment. Using this information, it is possible to divide the global navigation task into a sequence of landmarks which must be successively reached. The supervision algorithm then allows to select the right task at the right instant and to guarantee a smooth switch between the different motions. Our works have been validated by simulations and experimentations, demonstrating the efficiency of our approach.
