Design And Realization Of Robot Quick Reset Positioning System Based On VSLAM

Intelligent robots have gradually been integrated into our daily life and work.The structure of robots has become more and more diversified according to needs and functions: such as quadruped robots,which are widely used and have strong environmental adaptability.They are used in disaster relief and military counter-terrorism.Plays an important role.Compared with traditional wheeled robots,quadruped robots can flexibly perform actions such as climbing,ground support,and obstacle crossing.However,when facing indoor and densely sheltered environments such as basements,forests and other complex environments,it is prone to unstable or missing positioning signals such as the Global Position System(GPS).Most of the existing positioning solutions for quadruped robots are many It is mainly based on laser positioning,which is costly.The solution based on visual positioning and map construction(Visual Simultaneous Localization And Mapping,VSLAM)has the advantages of simple accessories and low cost.These advantages are very important for the future development and popularization of a technology.important.This article aims to design a positioning scheme that does not rely on external positioning information,and to study the visual positioning system of a quadruped robot based on VSLAM.This article first designs a VSLAM-based positioning method for quadruped robots.Based on the ORB-SLAM2 framework in VSLAM,a method for quick reset positioning is studied: by saving relevant feature points and key frames in parallel at runtime Harmony View realizes resetting and positioning of historical scenes;improving the data access algorithm,enabling data to be accessed in bin format,improving data access efficiency,thereby expanding the application scenarios of the quadruped robot positioning system at a lower cost.Secondly,in view of the low efficiency of the original visual odometer of the ORB-SLAM2 framework,this paper uses the three feature extraction methods of ORB,SURF,and SIFT in the common front-end visual odometry calculation method,as well as BF(Brute Force Matcher)and FLANN(Flann Based Matcher).The comparison of feature matching methods leads to the conclusion that the ORB algorithm and the FLANN algorithm are more suitable for the scenarios in this article.In this way,the improvement of the visual mileage calculation method can be achieved by calling the GPU to assist the extraction of feature points,thereby improving efficiency.Then,in view of the problem that ORB-SLAM2 cannot provide dense 3D maps,the map construction method is improved through the octree map,and the real-time dense 3D map construction algorithm is realized,which lays the foundation for further research on obstacle avoidance and navigation functions.The back-end optimization method based on the Bundle Adjustment method is used to achieve the global beam square error and the robust kernel function to enhance the robustness of the system.A loop detection strategy based on the bag-of-words model is adopted to improve the efficiency of loop detection.Finally,the accuracy of the positioning system,the efficiency of ORB feature extraction and matching,and the comprehensive performance of the equipment were verified on the quadruped robot.