Simultaneous Location And Mapping Of Mobile Robot Based On Semantic Segmentation

ORB-SLAM2 is a visual SLAM(Simultaneous Localization and Mapping)system based on ORB(Oriented FAST and Rotated BRIEF),a kind of image feature.It is one of the most perfect and practical systems in modern SLAM system,and Its advantages in real-time and accuracy have been concerned and recognized by many researchers.In this paper,the software flow and mathematical theory of ORB_SLAM2 system are analyzed and studied firstly.Then,based on TuretleBot2 mobile robot,the original ORB-SLAM2 system is improved in ROS:add semantics segmentation function to increase the stability of the system in dynamic scene is improved;add real-time mapping function to build three-dimensional map for obstacle avoidance and navigation.In this paper,the improved system is tested in the simulation environment and the actual scene respectively.The test results prove the superiority of the improved system.This paper has completed the following research work.(1)The software architecture and flow of ORB_SLAM2 system are analyzed,and the ORB feature algorithm,pose estimation,non-linear optimization,loop detection and other theories are analyzed and studied.At the same time,the shortcomings and improvements of ORB_SLAM2 system are analyzed,which lays a theoretical foundation for future research.(2)Aiming at the inaccurate pose estimation of ORB_SLAM2 system in dynamic scenes,a method is proposed to improve the stability of the system,which of eliminates dynamic feature points distributed on human and animals by semantics segmentation.The improved system has been tested on the common data set,and the test results show that the error of the improved system is obviously lower than that of the original system.(3)In order to build a three-dimensional map for robot navigation and obstacle avoidance,OctoMap mapping function was added to the original ORB_SLAM2 system.In order to ensure real-time and reduce the space occupied by the map,a strategy of selecting key frames of the map according to the number of common-view map points of key frames was proposed.After adding the mapping function,the system has been tested on the common data set,which shows good results in real-time and space occupancy.(4)In order to test the real-time and reliability of the improved system in the actual environment,this paper has tested the improved system in the actual environment on TurtleBot2 experimental platform with the help of ROS(Robot Operating System),has has shown good results.