Research And Implementation Of Multi-robot Visual SLAM Algorithm

How to detect and map the unknown environment without the help of other positioning systems(such as Global Positioning System)is of great practical significance,such as disaster scene,deep space exploration,even underwater or battlefield environment.Vision based Simultaneous Location and Mapping vSLAM)provides a powerful solution and has become a hot topic in autonomous robot.With the increase of task complexity,diversity and system stability,the visual SLAM of single robot has been challenged in the aspects of execution efficiency,detection range and system stability.Based on this,this thesis starts with the basic theory of single robot visual SLAM,studies the key points and difficulties of multi-robot visual SLAM,develops algorithm for multi-robot visual SLAM and conducts experiment to verify the proposed algorithm.Firstly,based on the theory and mathematical model of visual SLAM,the front end and back end of single robot visual SLAM algorithm are researched and implemented effectively.Included are feature selection algorithm based on minimum threshold,key frame selection algorithm,visual odometry,loop closure detection based on feature word bag method and global backend optimization algorithm based on graph optimization.In the aspect of mapping,based on binocular vision,the disparity is calculated by semi-global block matching algorithm,and then the depth of each pixel is determined by geometric principle.Further,the point cloud of the scene is constructed by filling point cloud with RGB and depth.After that the scene information is reconstructed.Secondly,aiming at the multi-robot SLAM problem,on one hand,this thesis uses the visual word bag and index of multi-robot system respectively to detect the common area among robots efficiently,the pose transformation and the coordinate unification between the robots are completed through the pose correlation and index between the robots and the precise calculation of the relative pose between the robots with the iterative closest point(ICP).On the other hand,a simplified multi-robot graph optimization pose model is established to improve the global optimization parameters and optimize the multi-robot back-end pose.According to the unified inter-robot coordinates and the updated global pose,the sub-map created by each robot is fused to the global map.So far,the global consistent robot coordinates and the more accurate global map are established.Finally,based on the established mobile robot system platform,location and mapping experiments of single robot and multi-robot are carried out.Experimental results verify the effectiveness of the proposed algorithm.