Research On Matching And Tracking Algorithms Of Associated Features For Visual SLAM

With the development of simultaneous localization and mapping(SLAM),the feature-based visual SLAM methods have been enriched and improved.As an important part of the visual SLAM technology,feature tracking is the basic step of the pose estimation and optimization,which affects positioning accuracy and mapping quality.This thesis studies the matching algorithms of the associated features.In this thesis,some reliable feature tracking algorithms are proposed.The main works of this thesis are listed as follow.(1)A feature points matching algorithm based on the geometric cluster pyramid is presented.The matching results satisfying both descriptor similarity and geometric constraints are obtained,which reduce the incorrect and missing matching problems.(2)An edge feature-based tracker is presented.The edge features are extracted from the subpixel edges and matched according to the consistency of the motion.The matching results of the edge features are refined by the geometric clusters.(3)The pose estimation methods based on the associated point and edge features are proposed.The poses of the cameras and the position of the landmarks are optimized.The effectiveness of the proposed tracking algorithms of the associated features for visual SLAM are verified by using both the public data set and the practical data collected by a robot experimental platform.The accurate and reliable matching results of the associated features can be achieved between the consecutive frames,which can meet the requirements of positioning and mapping in the visual SLAM.