Research On Indoor Robot Relocation Based On Image Features

Today,the rapid development of mobile robots has penetrated into every corner of society.Among them,SLAM(Simultaneous Localization and Mapping)is considered by many scholars as the key to solve the autonomous control of mobile robots.The integration of RGB-D vision sensor and SLAM algorithm is a major breakthrough in the field of SLAM research.This combination can make the robot have the ability to perceive the surrounding environment like human eyes do,and acquire the color information and depth information of the surrounding scene through the depth camera,thus construct a three-dimensional map of the environmental scene.The main work of the RGB-D visual SLAM algorithm based on image features in this article is as follows:(1)Construction of the SLAM map.Using TurtleBot to capture color images and depth images in environment scenes,extract features of color images and calculate their corresponding descriptors,and then perform feature matching,estimate motion transformation based on matching results and optimize them;finally,construct poses according to the optimized results.Figure,then closed-loop detection and pose diagram optimization,and finally build a map based on the requirements of post robot pose recovery.(2)Matching the similar images by using bag of words algorithms.The feature descriptors of the map set image are clustered and a visual dictionary is constructed.Then the NAO robot is used to enter the scene to acquire the image,and the image is preprocessed to construct the visual vocabulary,and then the similarity between the feature set and the map set image is calculated.Get the image with the highest similarity in the map set as the matching result.(3)Robot pose restoration.The constructed SLAM map is actually a collection of all the landmarks.Through the map we can obtain the keyframes and their corresponding poses.According to the results of image matching,the coordinates of the feature points are calculated,and the position and orientation changes of the two frames of images are estimated by the epipolar geometry principle,and eventually relocate robots entering the scene.Through the above process,the map constructed by the RGB-D SLAM algorithm canbe effectively utilized.The experimental results show that the proposed method can effectively locate the robots entering the map environment.