Realization Of SLAM Based On Improved ORB Key Frame Detection And Matching

In unknown environment,robot obtains the external environment information through sensors,incresmentally constructs and updates a map of the environment,and simultaneously estimates its pose in the global environment model,this process is the robot Simultaneous Localization and Mapping(SLAM).SLAM has been one of the most popular robotic research fields,it has theoretical significance and application values for robot task planning and autonomic controls.For sensors used in the SLAM,RGB-D camera is one kind of low-cost environment modeling sensors.It samples the environment color and depth information at the same time,which can greatly simplify the depth information computing process.Because of the noise in color and depth images and accumulated error in the depth parametric process,there are lots of uncertainty values in the process of estimating robot pose and mapping,which can cause the inconsistent pose estimation and map drift.The sense data with serious errors can generate an error environment model and global localization information.The Loop Closure Detection in the SLAM system can greatly reduce the map drift and enhance the consistency of the estimation pose.Around how to build a robust and real-time RGB-D SLAM system,proposed a Loop Closure Detection algorithm based on ORB features KeyFrame DoVW in SLAM,the main research work and achievements are as follows.a)Described the RGB-D sensor work principle and the basic calibration principles,the the joint sensor calibration method for the color and depth sensors was proposed.Zhang's method was used to correcting the distortion of the RGB camera and depth camera,while using joint alignment algorithm for the rgb and depth camera to obtain consistent image information from the same perspective;b)In this paper,we study the feature point detection in the SLAM front-end,analyzed and compared SIFT and FAST feature point detection algorithms,SIFT and BRIEF feature descriptor algorithms.According to the environmental characteristics and real-time requirements,we focus on ORB algorithm which based on the FAST and BRIEF algorithm,and enhance its rotational invariance and real-time performance;c)Introduced and studied the graph-based SLAM system.Frame-to-frame alignment,KeyFrame Loop Closure Detection and graph optimization are three main aspects in graph-based SLAM.Proposed a SLAM system based on imporoved ORB KeyFrame detection and matching algorithm;d)How the different algorthims affect the SLAM system were discussed,and established the relationships between the KeyFrame Loop Closure Detection algorithm and the SLAM system robustness.The improved SLAM system can be divided into three parts.The front-end constructs a pose graph,the back-end optimizes and updates the pose graph and the last part is map-expression.An improved ORB feature detection and matching algorithm was used to realizing a fast and effective matching between two adjacent RGB frames.According to the camera perspective projection model and dense image frames,the adjacent matched 2D frames can transform to 3D color point clouds.The relative pose between the adjacent frames was computed by improved RANSAC-ICP algorithm,which can solve the mobile robot precise localization problem.The Loop Closure Detection algorithm based on KeyFrame BoVW can improve the localization and mapping speed and consistency,reduce redundant model structure and generate a map with consistency;e)The real-time and robust performance of the improved SLAM system is evaluated by frame-to-frame matching speed and root-mean-square(RMSE)of the absolute trajectory error(ATE).Compard with the traditional RGB-D SLAM system,the standard test results indicate the effectiveness of the improved SLAM system.Finally,built an mobile robot experimental platform,and the field tests under different experimental environment verify the robustness and real-time performance of the improved SLAM...