Research On Binocular Vision Localization System For Indoor Robot Based On Natural Landmarks

Home service robot is the product of robots with artificial intelligence,which get gradually the favour of people.However,how to realize the accurate localization of the home service robot is still the most important problem of the family service The robot localization method based on vision is in full swing.Binocular stereo vision is widely used in robot localization system,which can obtain information of the environment and recover the depth information of the scene.This thesis studies the key technology involved in binocular vision localization system for indoor robot,the ceiling corner is not easy to change that as natural landmarks to achieve robot localization.Before the robot localization,it is need for the ceiling corner recognition,using the invariant SIFT descriptor to descript the features.In robot localization stage,using the 3D reconstruction principle and trilateration measurement principle to achieve localization.finally,completed software development and robot localization experiments on the Traveler II Robot platform.The work of this paper is summarized as follows:Firstly,this thesis introduces the binocular stereo vision system and principle of robot localization,and the Software implementation of functional modules is introduced,which lays the foundation for the following experiments.Secondly,This dissertation discusses feature extraction and stereo matching based on SIFT algorithm and FAST algorithm,considering the rapidity for FAST feature points extraction and robustness of the SIFT descriptor algorithm,for the application environment,raising the improvement of FAST algorithm and SIFT algorithm.A method that based on KSW entropy adaptive double threshold FAST extraction and subsectional SIFT descriptior stereo matching is improved.the feasibility and effective for proposed method are verified by the experiments.Thirdly,this thesis introduces a corner feature database method based on K-means clustering,classification storage of indoor corner feature database,to speed up the search speed of the corner,using SIFT stereo matching method to achieve the recognition of the corner,further realizing the localization of the indoor robot,and the localization accuracy of the robot is analyzed.Finally,this thesis depending on Microsoft Visual C++6.0 and Open CV software,realizing feature extraction and stereo matching algorithm,K-means clustering algorithm,the three-dimensional coordinates of the corner feature points as well as the threedimensional coordinates of the robot.It functions on the Voyager-II Robot experiment platform successfully.