Research On Feature Point Matching And Mapping Based On ORB-SLAM

Simultaneous Localization and Mapping(SLAM)is the mainstream research method in the field of computer vision and robotics.It refers to the robot start moving from any position in the unknown environment and then positioning according to the sensor data while constructing the surrounding environment.The SLAM with the camera as the only external sensor is called visual SLAM.Since the camera is low in price,light in weight,easy to equip on other hardware,and the image contains a wealth of information,visual SLAM has made great progress in recent years.ORB-SLAM is a versatile and accurate SLAM solution for Monocular,Stereo and RGB-D cameras.And ORB-SLAM is currently the state of the art visual SLAM algorithm.This paper analyzes the various modules in the ORB-SLAM system,expounds the principles and functions of each module and introduces some algorithms related to visual SLAM,which provides the necessary theoretical basis for subsequent research.As more and more projects are applying SLAM to smart phones and embedded devices,the speed of algorithms becomes more and more important.In feature matching algorithms,different feature detection algorithms and feature description algorithms have a great influence on the speed of the system.In this paper,the feature detection algorithm is studied,and the commonly used feature detection algorithm is introduced in detail.The superiority of ORB algorithm is verified by experiments.To improve the speed of the algorithm,this paper proposes the fusion of ORB algorithm and AGAST algorithm,and proposes an improved feature matching algorithm based on ORB-SLAM system.The experimental results on the public dataset show that the proposed method improves the speed of the algorithm while keeping the robustness of the system.The positioning of the robot is the main task of the visual SLAM,and building the map is another important task.The map is an important bridge between the SLAM system and the upper application.It is very important for the operator to visualize the map of the remote robot.Offline construction of the visual map greatly reduces the convenience of operation.ORB-SLAM can build sparse feature point maps in real time,but this type of map is not intuitive enough for the operator.In this paper,the map construction in visual SLAM is deeply studied,and the commonly used map forms and related algorithms for constructing point clouds are introduced.This paper proposes a scheme for constructing a dense point cloud map in real time,and optimizes the map through voxel filtering algorithms.This paper also proposes a scheme for constructing an octree map in real time,which further reduces the occupied space of the map.The experimental results on the public dataset show that the proposed scheme is effective.