The Accurate And Efficient Monocular Simultaneous Localization And Mapping Algorithm Research

Simultaneous Localization And Mapping is considered to be the key to realization robot autonomous intelligence,and it provides support for the interaction between the robot and the environment.In recent years,with the rapid development of Driverless cars,Augmented Reality,Smart Cities and other fields,making SLAM research more and more hot.Among the many SLAM solutions,monocular vision sensors are sought after by researchers due to their low cost,rich information acquisition,compactness and convenience.Therefore,SLAM technology based on monocular camera has become a popular direction for all research institutes,and is of great significance for realizing the AI and digital age.Due to the complexity of the environment itself,coupled with the fact that visual SLAM involves a wide range of research fields,research still faces many difficulties.In this article,I fully and in-depth studied the existing outstanding solutions,compared the advantages and disadvantages of the program,take full account of the camera movement mode,propose a mapbuilding method based on rotation,analyzed the loopback detection model,made a more accurate and fast loop detection algorithm.The main work and innovation are as follows:In this work,I studied the key technologies in monocular vision SLAM,such as camera imaging model,camera movement patterns and mathematical representations,image feature extraction matching,map initialization,obtaining of 3D Coordinates of Camera Pose and Map Point,Graph Optimization Algorithm and so on.And I compared different implementations of the existing SLAM solution and identified the overall SLAM framework program.I analyzed the local map construction method and summarized the keyframe selection goal.In the proposed system,a key frame selection strategy based on rotation degree is proposed to solve the problem that original key frame selection scheme can not add key frames in time when the camera rotates to a large extent.This work gives the definition and calculation of rotation degree and experiments.The system makes full use of the rotation degree to judge the motion of the camera.When a large rotation occurs,the system first adds the key frame with a loose strategy and then removes the redundant key frame after completing the local map construction.I validated the algorithm on a public data set,and the experimental results show that the algorithm is more accurate and robust.In view of the low efficiency of loopback detection,a new algorithm of loopback detection based on historical model is proposed.By taking advantage of the data at the time of the history loopback,it is possible to predict the possible location of the next loopback,using skip and frame-by-loop loopback detection in different confidence intervals.After the loopback loop detection detects a suspected loopback,the policy is adjusted in time to re-detect from the skipped frame.Experiments were performed on a standard dataset,the hop length was obtained,the real-time performance and accuracy of the algorithm were verified,and the efficiency of loopback detection was improved.An improved monocular SLAM is used to conduct experiments in laboratory indoor environment,respectively,to obtain an accurate map in real time.