Research On Self Localization Of Mobile Robot Based On Vision

With the rapid development of mobile robot technology,many areas of the application of the technology are requested more widely.The indoor environment of the mobile robot is gradually moving into everyone's field of vision.With the development of intelligent mobile robot,in addition to the already mature outdoor localization technology,more and more research methods are indoor localization However,there are some problems in the self-localization of indoor mobile robots,such as low accuracy of localization,poor real-time localization,large amount of data and complicated calculation.In this paper kinect sensors used to capture the indoor environment of mobile robot three-dimensional information,and the self-localization of mobile robots is carried out by using SIFT(Scale-invariant feature transform)image matching algorithm and Iterative Corresponding Point(ICP)algorithm based on feature points.The main work is as follows:1)In the indoor environment,Kinect sensors are used to acquire image information.Kinect get color image and depth image information of different data frames.Through the Kinect calibration,the 3D point cloud data can be generated rapidly,which realizes the combination of the two-dimensional color data and the depth data.The method is based on Kinect's own toolkit and algorithm,and the accuracy of the data is higher.2)Color images and depth information of two consecutive frames are obtained by using kinect sensor.Feature points of color images are extracted by SIFT algorithm.Then,fast matching of feature points is performed by K_D tree neighbor search.RANSAC algorithm is used to remove the wrong matching points.After matching,the corresponding depth information is found,and the feature points and depth information are transformed into three-dimensional data.The experiments results illustrate that the accuracy of the detected feature points has been highly proved and there is lower probability of the error matching points after using the improved algorithm.3)In order to overcome the large amount of data acquisition and time complexity in the process of moving,We decided to use the feature point-based ICP algorithm,we used the motion estimation method,the position and direction of the camera with better robustness can be obtained finally,while obtain the rotation matrix and translation vector between two adjacent frames.In the process of robot movement,the parameters are updated,so that the robot can achieve the purpose of updating the position and position.The position error of the robot pose information in the horizontal direction is about 9.237cm,and the positioning error is about 5~0 in the horizontal and vertical viewing angles.This method has the feasibility.