Research On Mapping Technology For ORB-SLAM

In the field of mobile robot,SLAM(Simultaneous Localization and Mapping)has become an increasingly popular research topic in the past 30 years.As an important content of SLAM technology,mapping has been researched and paid attention by scholars at home and abroad.Different scenes need different forms of maps.With the development of computer vision and SLAM technology,3D Mapping is widely used because of their rich information.ORB(ORiented Brief)-SLAM,as a perfect,easy-to-use,modern visual SLAM system,has been developed and used by many researchers,However there are still defects in that the original map is sparse,and the map could not be saved.The main work of the paper is as follows:1.In the ORB-SLAM system,BA(Bundle Adjustment)plays an important role.BA optimization is essentially a nonlinear least squares problem.The ORB-SLAM system is optimized by the Levenberg-Marquardt method,which has redundancy in the linear equation,slow convergence and low efficiency.This paper introduces a sparse BFGS(Broyden?Fletcher?Goldfarb?Shanno)nonlinear optimization method for solving the BA problem,which combines the Gauss-Newton method and the Levenberg-Marquardt method to approximate Hessian matrix.A gain matrix is calculated by using both the Jacobian matrix and residual vector,which results in a better estimation of the descent direction and step size,fewer iterations and better efficiency.2.Due to the importance of the 3D mapping,and the sparse map constructed by the ORB-SLAM is not practical enough.Firstly,the paper uses Kinect to obtain the color images and depth images of the environment and then fuses the keyframes of ORB-SLAM to generate dense point cloud map and realizes localized preservation of dense point cloud map.Then,the point cloud map contains rich information,which leads to redundancy.The paper uses the Octomap to realize the compression and multi-resolution representation of the map.Finally,a geometric consistent plane extraction method is introduced to segment the scene according to orthogonal and parallel constraints of 3D scene,which results in constructing structural 3D point cloud map and then a hierarchical Octomap is constructed.In this paper,the ORB-SLAM system is optimized from the perspective of BA optimization and mapping.The experimental results show that the sparse BFGS-based BA optimization algorithm is faster and more robust than the BA algorithm in the ORB-SLAM.In terms of mapping,3D dense map,Octomap,and hierarchical Octomap are verified in the real environment.