Research On Vision Based Simultaneous Localization And Mapping With Multi-Robot System

With the continuous development of intelligent society,the demand for intelligent robots in various industries is getting higher and higher,which also brings a long-term development to Simultaneous Localization and Mapping(SLAM)technology.At present,SLAM algorithms based on various sensors have made significant progress,especially based on the visual sensor SLAM solution has been widely used in practice.It can be found that the development of single robot SLAM technology is constantly maturing,and the research direction in this field is also extended to the multi-robot collaborative SLAM technology.Today,with the continuous development of SLAM algorithm,this paper studies and experiments on the vision-based multi-robot cooperative SLAM solution.This paper first expounds the development status of SLAM,focuses on the analysis of the principle of SLAM problem,and introduces the SLAM solution based on different sensors.Then the vision-based SLAM problem solution is analyzed from three aspects: the application principle of the camera,the principle of visual odometer and the principle of back-end optimization.The ORB(Oriented FAST and Rotated BRIEF)feature was chosen as the basis for the visual SLAM solution.On this basis,the implementation principle of ORB-SLAM algorithm is analyzed in detail,and the sparse point cloud map generated by this scheme is selected as the basis of subsequent research.Compared with the current multi-robot collaborative mapping technology,it is mainly proposed for grid maps.Compared with the point cloud map selected in this paper,the grid map is not rich enough for the environment,and the map fusion is difficult.Cloud maps can determine map similarity through image matching,and online map fusion relies on accurate recognition and location between robots.In view of the above difficulties,this paper proposes a multi-robot collaborative mapping solution.The scheme mainly includes three aspects.Firstly,a multi-machine communication scheme based on ROS(Robot Operation System)is designed.Then,when the robots enter each other's field of view,a target detection scheme based on deep convolutional neural network is designed.Finally,the robot is successfully identified.The sub-maps of the robot members are merged.In the experimental part of this paper,the ORB-SLAM algorithm of single robot is firstly tested on the data set and the actual environment.The real-time performance of the algorithm is proved by the experimental results.The sparse point cloud map constructed in the experiment provides the possibility of real-time fusion of multi-robot maps.For the experiment and debugging of multi-robot collaborative mapping,comparing with the related schemes under the data set,it proves that the proposed scheme has certain superiority,and at the same time,the proposed scheme is tested in the real environment.The purpose of this experiment is to verify the feasibility of the scheme and compare it with the map established by a single robot.It proves that the proposed algorithm has certain advantages in robot positioning and map rendering efficiency.The experimental results show that the map fusion effect is roughly in line with the expected target.