Research On Visual Simultaneous Localization And Mapping Based On ROS

SLAM is mainly used to solve the problem of map construction based on the sensor information of mobile robot in unknown environment and the positioning problem caused by the error in the process of moving.Visual SLAM is a system which uses camera as sensor to collect data.Based on RGB-D SLAM algorithm,some modules are improved in this paper.Experimental results show that the algorithm can improve the accuracy of the system.The thesis mainly includes the following parts.(1)In the front-end part,we focus on the matching problem of image ORB feature points.According to the problem that the accuracy of feature point matching is not high,the location information of feature points on the picture is introduced to eliminate mismatches.Experiments show that the improved algorithm can improve the accuracy of matching.In the aspect of pose estimation,the local feature point map is compared with the current frame to solve the problem that the depth value may be inaccurate.(2)It focuses on the back-end links of the RGB-D SLAM algorithm,and proposes an improved key frame extraction algorithm based on the traditional key frame selection method,combining time,space and picture information.Experiments show that the improved algorithm can improve the accuracy of key frames.At the same time,a bag of words is used for closed-loop detection to detect whether there is a loopback problem,and finally the camera pose is optimized through the pose map to construct a global map.(3)Construct a ROS framework based on RGB-D SLAM,use standard data sets for simulation experiments,and evaluate the system through absolute trajectory error ATE.Experiments show that the improved system can improve the accuracy of SLAM.