Mobile Robot Global Positioning System Based On Computer Vision Research

In mobile robot applications,it is a fundamental and Important requirement that the robot should be able to localize itself accurately within its operating environments.It is a challenging research topic in mobile robot which has attracted many researchers.This thesis focuses on the study of mobile robot localization,and proposes a scheme based on vision navigation technology.The scheme implements mobile robot global localization with monocular vision in unknown environment.In consideration of time-consumed,the project is mainly applied in correcting errors of the main navigation system of the mobile robot in indoor environment.Mobile robot localization is based on the map of its environment.In contrast,the map building of the environment is based on the accurate localization of the robot.In unknown environments,this is a conflicting and correlated process.This article deals with two kinds of different environment:known and unknown.When a mobile robot enters a room,it first judges whether the room is familiar or not.If not,an investigation procedure is carried out and an electronic map is developed.Then the global localization can be realized with the methodology used in a known environment.At low-level processing of vision aspect,the SIFT features given by Lowe are used to act as the main characteristic to build environment map.The SIFT features are highly distinctive,and invariant to image scale and rotation,and are shown to provide robust matching across a substantial range of affine distortion,change in 3D viewpoint,addition of noise,and changes in illumination,so they can be used to describe the environment characters correctly.While building electronic map for indoor environment,the pictures of environment are obtained from several viewpoints,and the targets are computed with 3D-Reconstruction of double views.For several results of a same target,two type approaches based respectively on assumption of independence and assumption of correlation are used.Meanwhile,the RANSAC method is used to enhance precision and stability of the result.For mobile robot localization,according to relative relationships between viewpoints and correspondences of environment map,the result is calculated directly with the theory of camera perspective projection.The feature matching is implemented with the KD-Tree-based nearest search approach.Because of outstanding search efficiency,the mobile robot takes up less time.Similarly,the RANSAC method is also used to process the result.