Research On Relocation Algorithms In Multiple Scenes Based On Image Features

With the rapid development of intelligent mobile robots,it has become very important for robots to have the ability to autonomously locate and establish map tasks.Simultaneous Localization And Mapping(SLAM)technology has become the focus of robotics research.The current SLAM systems have been able to achieve real-time mapping of surrounding scenes.However,SLAMs cannot be performed in large-scale scenarios and many documents do not specify how to use these maps.Therefore,SLAMs and multiplexed SLAMs are performed in large scenes.Maps have become an urgent issue to be solved in the field of research on intelligent mobile robots.Through the analysis of the existing SLAM technology,the drawbacks of traditional SLAM technology in large-scale scenarios are highlighted.The existing SLAM technology and visual vocabulary tree algorithms are introduced and analyzed,and their advantages and disadvantages are summarized.The existing SLAM technology is improved:(1)Divide the large scene into multiple sub-scenes.Use the RGB-D SLAM to construct the scene.Use the visual odometer to estimate the robot motion and establish the pose diagram.Use the graph optimization principle to globally align the pose diagram.Optimize and build a topological map based on multiple sub-scenarios.(2)This paper introduces the use of the word co-occurrence matrix to describe the scene,and derives the similarity measure of the word co-occurrence matrix from the perspective of the graph kernel.Thus,the relocation process is divided into two phases.The scene is identified first,and then the topology map in the scene is repositioned.(3)Taking into account that the scene information contained in one image may not be much,so this paper proposes to use Bayesian filtering for the multi-image scene similarity result and the filtering result as the result of scene recognition,thereby improving the scene recognition efficiency.