Research On ORB-SLAM Algorithm For Indoor Environment

The intelligent robot lacks GPS signal in the Shadowed or semi-shadowed environment,so it needs to judge its own position through the perceptive navigation algorithm,SLAM(Simultaneous Localization and Mapping)can effectively solve the problem of location perception for unfamiliar scenes.Visual SLAM uses the camera as a sensor to obtain the 2D image data of the scene,and then calculates the pose changes according to the motion model and observation model,maps the 2D points in the image to the 3D space to obtain the point cloud map of the scene,and conducts positioning and navigation based on the point cloud map.The main research work of this paper is as follows:(1)This paper analyzes the front end,back end and loop closure of the SLAM algorithm,and obtain its functions in the SLAM algorithm,and analyze and explain the influence of each module on the mapping accuracy,illumination robustness and other performance indexes of SLAM.(2)This paper analyzes and improve the sparse feature method for SLAM.The selection of feature descriptors has a great influence on the visual SLAM algorithm.This paper analyzes ORB descriptors.Due to the assumption of grayscale consistency,the algorithm cannot cope with harsh lighting conditions when running,so the algorithm is not robust.In this regard,the H component's intensity illumination robustness in HSV space is used to calculate the H component histogram in the neighborhood through the Gaussian Mixture Model,and then the feature point matching is carried out to improve the illumination intensity robustness.(3)This paper summarizes the research method of the back-end of SLAM algorithm.The advantages and disadvantages of linear optimization algorithm and nonlinear optimization algorithm are compared.The linear optimization algorithm assumes that the motion state is Markov,that is,the current state is only constrained by the previous state,and all the states before the previous state are ignored,so there is a certain error.The nonlinear optimization algorithm based on global constraints,considering the global constraints information,makes the optimization effect more accurate,but the calculation pressure is large.By means of elimination,the sparse of determinant is used to improve the real-time performance.(4)This paper summarizes the structure of loop closure based on BoW algorithm.It is found that the method of loop matching which traverses all the historical frames and calculates their similarity with the current frame has a low computational time efficiency.In this regard,a loop detection algorithm based on scene flow clustering algorithm is proposed:since the BoW similarity of scene frames in scene flow conforms to the Gaussian distribution,the semantic information of scene flow is used to cluster them,so that the scene flow is segmented and several different scenes are obtained.The scene to which the current frame belongs can be obtained to avoid redundant loop frame calculation.Then,Dense SIFT descriptors can be used to accurately match loop frames in the scene to improve the efficiency and accuracy of loop closure detection.To sum up,the proposed algorithm has been improved in terms of illumination robustness,SLAM mapping accuracy and real-time performance,which is of positive significance for the practical application of SLAM algorithm.