Research Of SLAM Method Based On Monocular Vision

Simultaneous Localization and Mapping under unknown environment is one of the most challenging problems in the field of research on current mobile robot localization and navigation. The related researches of SLAM issues have important theoretical significance and practical value to achieve a high degree of autonomous and intelligent for robot. In recent years, with the development of the computer vision technology and the wide application of vision sensors, the research of mono-SLAM technology has gradually become an important research direction in SLAM.There are several main difficulties in researching of SLAM based on monocular vision information.(1)How to extract stable feature points in limited number and uniform distribution.(2)How to get features’ depth information and complete its initialization.(3)How to build accurate model of motion and observation about monocular camera.(4)how to implement the dynamic management of map features.etc. According to the first question, here carried out research on the extraction and matching of scene feature based on SIFT algorithm. When explored characteristics of SIFT feature points and analyzed the situation of its matching between different frames, We know that SIFT algorithm can extract rich and stable feature points. In order to limit map features within finite number and uniform distribution, A method of SIFT extraction based on significant parameter index of local area is put forward. For the second question, here proposes a non-lazy initialization method combined with reverse depth estimation. After initialization, the information of feature points is joined to the full state vector directly for EKF estimation and update. As to the last two problem, combined with the frame of EKF, a monocular EKF-SLAM algorithm based on SIFT feature is proposed after the establishment of full state model and observation model. The corresponding operating procedures or treatments of system initialization, EKF state’s estimation and update,feature map’s creation and dynamic management are given. Besides, this article also designs a process about mono-EKF-SLAM simulation, and verifies the feasibility of the proposed algorithm in MATLAB.Final results show that, using improved feature, not only feature points’ number is moderate, but its spatial dispersion is well. When used in SLAM algorithm, it has higher convergence rate, and smaller final estimation error. It proves that the presented method can finish the task well to create the map of indoor unknown static scene under the condition of only using a single camera. It not only can save the cost,but obtain a satisfactory accuracy.