Research On Multi-Robot Collaborative Mapping Based On Visual-Depth Information

Simultaneous Localization and Mapping(Simultaneous Localization and Mapping,SLAM)is a technology that realizes Simultaneous Localization and pose estimation of mobile robots in unknown environments and constructs environment maps through selfcarried sensors.With the rapid development of this technology,it has been widely used in many fields such as daily life,rescue and disaster relief,national defense industry and so on.At present,many mature solutions have been developed to solve the problems related to the simultaneous localization and mapping of mobile robots.However,there are still many difficult problems to solve when the robot explores a large scale scene independently for a long time.Therefore,multi-robot cooperative mapping has become one of the current hot research topics.The synergistic effect between robots can not only improve the efficiency of location mapping,but also enhance the intelligence and robustness of the whole system.The robots in the multi-robot cooperative mapping system not only need to complete their own SLAM process,but also need to transmit the acquired environmental information,determine the overlapping area of the local sub-map and fuse the map,and finally get a consistent global3 D map for sharing.In this paper,we mainly study the visual depth information based SLAM algorithm,the overlap region determination between local submaps and the fusion algorithm of 3D dense maps.The main research contents of this paper are as follows:(1)The framework of SLAM method based on RGB-D information is studied,and four modules of front-end visual odometer,back-end optimization,loop detection and map construction are analyzed respectively,and each module is theoretically deduced.First by RGB-D camera acquisition unknown scenarios of color and depth information,the front-end visual odometer the ORB algorithm based on feature point method of feature extraction and matching between image frames,and uses the iterative closest point(ICP)method to solve the camera motion between adjacent images,the back-end optimization method was optimized by using posture figure,loopback detection part through the word bag model(Bo W)eliminate the drift error,finally generate dense 3D point cloud maps.(2)Aiming at the overlapping region determination problem between local robot submaps,this paper designs an efficient determination method based on encounter recognition.Firstly,YOLO-V3 algorithm based on deep convolutional neural network is used as the robot recognition scheme to judge whether the two robots meet.If the encounter occurs,it can be clear that there is overlap between local maps,and the map transmission and fusion can be carried out.If not,the key frame transmission is carried out,and the visual word bag method is used to determine the overlapping area between local maps based on similarity judgment.This overlapping region determination method has good real-time performance and robustness,and defines the moment of local map transmission and fusion between robots.(3)Aiming at the problems of poor real-time performance and low accuracy of traditional 3D map fusion algorithms,this paper proposes a 3D map fusion algorithm based on low dimensional local feature descriptor.Firstly,Harris 3D algorithm was used to extract the key points,and then designs a histogram(SDH)based on spatial distribution of 3D feature descriptor coding feature information around the key,the Kd-Tree algorithm for radial feature matching nearest neighbor search is complete,use RANSAC algorithm is optimized,finally3D-ICP algorithm was used to achieve the 3D map fusion.Experiments show that the low dimension local feature descriptor designed in this paper has good description and superior rapidness,and the 3D map fusion based on this descriptor has good fusion effect,which can meet the accuracy requirements of practical application.(4)In this paper,the overall design of the multi-robot collaborative mapping system is completed under Ubuntu system,and the algorithm proposed in this paper is verified and tested experimentally.Firstly,the feasibility of the RGB-D SLAM system constructed in this paper is verified and its performance is tested.Then,the overlapping judgment method is verified,that is,the mutual recognition algorithm between robots.Then,the performance of the proposed 3D map fusion algorithm based on the low dimensional local feature descriptor is tested,and compared with the SHOT descriptor,the real-time performance and accuracy of the proposed algorithm are verified.Finally,the whole system is verified,which can fulfill the task requirements of multi-robot cooperative mapping,and obtain a consistent global 3D map.