In the context of outdoor mobile robotics, concepts of localization and perception are central to any achievement. Also, the work in this thesis improves an existing localization process more robust without significantly increasing their complexity. The proposed problematic addresses a robot in a potentially dangerous field with the aim to follow a path established as safe with a map as simple as possible. In addition, strong constraints are imposed as in the realization (inexpensive system, undetectable) as in the result (a real-time process execution and a localization continuously within a tolerance of 10 cm closed to the reference trajectory). The exteroceptive sensor chosen to carry this project is a camera while the pose estimation of the vehicle at each moment is achieved with an Extended Kalman filter. The main estimation problems are due to the non-linearity of the models and contributions provide some solutions: - an exact calculation method of the propagation of uncertainties in space world into space sensor (camera); - a method to detect the main event of a divergence of the update step of the Kalman filter; - a method to correct the Kalman gain. This project had two objectives: to achieve a localization function with respect to strong constraints previously mentioned, and allow a vehicle to leave temporarily the reference trajectory, while the operator modify the robot trajectory and then resume the normal course of its mission to the reference path. This second part involves a broader context in which it is necessary, in addition to the localization, to map the environment. This problem, identi ed by the acronym SLAM (Simultaneous Localization And Mapping), made the connection with the last two contributions of this thesis work: - an initialization method of the points which will constitute the SLAM map; - a method to maintain consistency between the reference map and the SLAM map. Results on real data, supporting each contribution, are presented and illustrate the realization of the two main objectives.
