Research On Key Technology Of Visual Simultaneous Localization And Mapping Based Class Specific Hyper Graph

With the development of artificial intelligence technology and the need of people’s life,the operating environment of a robot is becoming more and more extensive and complex,and even unknown.This characteristic determines that the robot(mobile robot)must be able to move independently in the environment which it detects,and build a consistent map of the environment,further locate itself according to the map to achieve real-time positioning,So then realize autonomous intelligence.The Localization and Mapping Simultaneous(SLAM)technique is proposed,which means a robot can estimate its poses and build a consistent map in the process of exploring the environment.SLAM has become the basis technology for the entire mobile robot autonomous navigation.Researchers around the world has taken their attention on it for nearly 20 years,and proposed a variety of solutions.In this paper,we study the monocular visual SLAM of the mobile robot based on the appearance of images taken from the unknown environment.The main work and innovations of this paper are as follows:1)We propose a loop closure detection method based on Hypergraph Model.Loop-closure detection is an essential part of SLAM,and the robot must be able to perform loop-closure detection in long-term,real-time position in the environment.n this paper,according to the similarity propagation,a Class Specific Hyper Graph is constructed by the attribute graphs represented the images of a range of scene.Based on the recursive SOM tree from the SLAM database,fast and accurately search loop-closure candidates,and then determine whether there is an occurrence of closed-loop according to the three constraints.2)A motion estimation algorithm based on optical flow is proposed.Motion estimation is the basis of SLAM.This paper location the features for initialization based on the Fast corners in the monocular visual position framework.On the basis of initialization,the following images are estimated directly according to the information of the image.The two times of Newton Gauss optimization and reprojection operation improve the accuracy of poses estimation.3)Combine the loop-closure detection based on hypergraph with a fast motion estimation,a monocular visial SLAM framework is constructed.Experimental results on the data set show that the proposed method is faster and more suitable for the single camera visual SLAM.The result is more accurate.