| An autonomous guiding system must be able to plan a task and execute it using its knowledge of the environment. The implementation of an autonomous parking system for an automobile vehicle in an urban environment implies the resolution of different problems: 3D perception of the environment, vehicle localization with respect to the environment, path planning, and path following and position control.;In our implementation, range data are used for the identification of a free parking place in the environment. A 3D representation of the parking profile is computed from a sequence of images of the global environment. A matching strategy using the phase of the image's Fourier transform produces a high density disparity map for each consecutive pair of images. A non-linear structure and motion estimation procedure is then used to generate the local depth maps. These local maps are integrated into an extended representation of the environment, in which a free parking place is detected.;Localization of the vehicle in the environment is accomplished by matching the local depth maps with the global parking profile model. Two alternative methods have been studied for path planning: the cell decomposition method, and the canonical path method. For the path following task, motion controllers have been developed based on a simplified dynamic model of the vehicle.;For experimentation, a reduced-size (scale 1:5) vehicle has been developed. This automobile-type vehicle can reach a maximum speed of 3 m/s. |
