Research On Multi-robot SLAM Based On RGB-D Camera

Simultaneous Localization and Mapping(SLAM)is the basis of robot intelligence.Nowadays,many excellent methods have been built to solve single-robot SLAM related problems and make it widely applied into reality.But single-robot SLAM still has many shortcomings in its application to the cases with larger scenes and longer tracking time.However,Multi-robot collaborated SLAM can increase the efficiency of robot mapping and make it more efficient.Compared with single-robot SLAM,multi-robot SLAM not only needs to complete the process of single-robot SLAM,but also needs to transfer data and integrate maps.Therefore,this paper mainly studies the problems related to RGB-D camera based single-robot SLAM,multi-robot data transmission and 3D map merging.The main research contents of this paper are as follows:(1)For the poor adaptability of feature point method applied to weak texture scene in the vision odometry of single-robot SLAM,causing target losing in tracking,direct method is added to the base of feature point method in this paper.This algorithm can select the appropriate method in vision odometry self-adaptively according to the number of feature points,which improves system robustness in complex scenes.In this paper,the effectiveness of the algorithm is verified on the TUM datasets,and compared with the visual odometry based on the feature point method,experimental results prove that our algorithm has higher stability and stronger adaptability in complex scenes.(2)For the low efficiency of data transmission method of the traditional multi-robot system,which takes up a lot of communication resources and is not conducive for its expansion,this paper proposes an efficient data transmission method based on key frames and descriptors.This algorithm introduces the key frame mechanism to determine the time interval of map transmission,which avoids the transmission of a large number of redundant data and improves the transmission efficiency.Meanwhile,the pre-transmission descriptor is used to determine whether the scanning areas of the two robots overlapped.The map will be transmitted only when there has an overlapping area,otherwise the map will not be transmitted.This strategy can effectively save communication resources and improve the scalability of the multi-robot system.The experimental results show that the multi-robot system using the proposed algorithm has a much smaller volume of total transferred map data than the traditional algorithm,and the time interval of map transmission is more reasonable.(3)For the low efficiency of traditional 3D map merging algorithms,this paper proposes a 3D map merging algorithm based on feature extraction and matching.On one hand,the algorithm uses calculated map normals to extract the feature points to reduce the computation amount of fusion;on the other hand,it puts forward the local descriptor based on the local reference frame and the reduced dimension to shorten the computation time.The experimental results show that compared with the traditional feature extraction algorithm,the proposed algorithm improves the matching efficiency while ensuring the correct matching rate.(4)Finally,the performance of the proposed algorithm is tested on the data sets.The robustness of the single-robot SLAM algorithm in weak texture scene is verified.Then the data transmission algorithm in this paper is tested and compared with the traditional algorithm to verify the transmission efficiency and intelligence of the proposed algorithm.Then the feature extraction and matching algorithm in this paper is tested and compared with the traditional algorithm to verify the robustness and real-time performance of the proposed algorithm.Finally,the whole system is tested,and the 3D map with global consistency can be obtained.