Research And Application On VSLAM Algorithm Based On RGB-D

Simultaneous Localization And Mapping(SLAM)has the advantages of high accuracy,robustness and consistency.The widespread application of RGBD cameras has made RGBD-SLAM one of the research hotspots in the field of mobile robot intelligence..The traditional RGB-DSLAM algorithm is divided into front-end and back-end.The front-end image information and depth information obtained from the RGBD camera are used to extract and match feature points based on the continuous images collected by the sensor.It is optimized pose information.Based on the above background,this thesis proposes the following methods for the problems of existing algorithms.The main research content is divided into three parts:(1)A rough-to-fine feature matching method based on Brute Force(BF)and Progressive Sample Consensus(PROSAC)is proposed to reduce mismatching caused by interference.A motion estimation algorithm based on hybrid iterative close point(ICP)and generalized-ICP(GICP)strategies was established to estimate the robot's motion trajectory,and the initial pose map was constructed.(2)An improved key frame selection strategy,key frame-based loop detection strategy,and map construction are proposed.The key frame threshold and DBoW3 bag-of-words model are used to complete key frame-based loopback detection.Use the L-M method for the pose transformation matrix.Construct an octree map that can be applied to mobile robot navigation.(3)In the mobile robot environment,the TUM standard test data set was used to test and evaluate the algorithms before and after improvement,and the absolute pose error(APE)index was used to test the experimental results and real camera positions.The posture data is analyzed and compared.The experimental results show that the algorithm in this thesis optimizes the RGBD-SLAM framework,and improves the efficiency and robustness of SLAM.