Research On Improved SLAM Algorithm Based On RGB-D

SLAM(Simultaneous Localization and Mapping)is recognized as the core-sector to realize the autonomous navigation of robots,which is also one of the most challenging subject.The RGB-D sensor,represented by Kinect,can obtain the depth information directly while acquiring the color information in the surrounding environment,the data processing process of which is simple and suitable for 3D map reconstruction.The SLAM study based on RGB-D sensor is called RGB-D SLAM,which is a hot topic in the field of robot autonomous navigation.In order to solve the problem of low efficiency and large error in the original RGB-D SLAM algorithm,this paper has researched respectively in improvement of the front and back end of the algorithm,and realized RGB-D SLAM system framework with high accuracy,robustness and real-time.Specific research results are as follows:Firstly,the study is used to collect the working principle,the internal and external parameters and the calibration method of Kinect for RGB-D information in the surrounding environment.By mean of the binocular joint calibration kit in MATLAB,the Kinect color lens and the depth lens is calibrated and aligned,and the point cloud image before and after calibration alignment is compared and analyzed to verify the calibration,which will contribute to improve the correct match rate for the image pixel of the RGB image pixels and Depth.Secondly,based on the several aspects of the front end of RGB-D SLAM algorithm,namely feature detection and descriptor extraction,feature matching,error matching culling,motion transformation estimation and motion transformation optimization,the improved error matching algorithm is proposed to combine the two-phase matching method with the threshold method,which takes less time,such as reduced by14.3%,14.7%,and 58.6% with the SIFT,SURF,and ORB algorithms,respectively,while the more correct number of matching points is preserved,such as increased by 5.7%,34.7% and 26.9% with the SIFT,SURF and ORB algorithms,respectively.Thirdly,based on the several aspects of the back end of RGB-D SLAM algorithm,namely the generation of the pose pattern,the closed-loop detection,the optimization of the pose image,the trajectory of the motion and the generation of the 3D point cloud map,the improved closed-loop detection algorithm is proposed which combine Close-range frame-by-frame closed-loop detection,long-range random closed-loop detection,and idea of BoVW,the bitmap generated by which is more neat,and the consistency is relatively good.Finally,The RGB-D SLAM algorithm before and after the improvement is evaluated by the open data set Computer Vision Group and the corresponding result evaluation tool,by which has been verified that the improved RGB-D SLAM system has higher accuracy and realism in constructing the map.In addition,the Turtlebot robot is equipped with Kinect for field experiment,and the continuously updated pose and 3D point cloud map can be generated with the system in the process of robot operation,and the robustness and effectiveness of the improved RGB-D SLAM algorithm has been verified.