Research On Key Technologies Of SLAM For Monocular Vision

Monocular vision SLAM has a lot of advantages,such as equipment cost is cheap,assembly process is easy and it can solve the problem that other binocular SLAM cannot Estimate the distance of infinity landmarks.So this type of SLAM in recent year has been received extensive research.After research the framework of SLAM,methods and principles monocular vision SLAM algorithm processes,the article determine to research critical steps based on monocular vision SLAM.After the study and implement traditional monocular vision-based SLAM,we improve this algorithm based on monocular vision SLAM algorithm.Experiments show that the improved been to this article SLAM system in real-time aspects of a marked improvement over the original system.In this paper,specific research work mainly includes the following aspects:(1)Describes the research and development status monocular SLAM methods,explains the basic principles of monocular SLAM problem and analysis an important realization of monocular vision SLAM method.(2)According to the requirement of SLAM and the principle of SLAM initialization,we implement a new initialization indicator of SLAM.This implement enhance the usability of this SLAM system.For SLAM system initialization,this indicator measures the number and density of landmark points and the direction of movement.This SLAM initialization method provides an effective support to SLAM system.(3)In SLAM area,filter malfunction caused by inconspicuous features in gray image may result in imprecise filter state estimation or even go to collapse.According to this,on the basis of multi-band image,we propose an algorithm which based on the federal Kalman filter to fuse position and orientation information.This algorithm uses the idea of decentralized filter that fusing the position and orientation information to obtain globally optimal state which will be re-allocated to each sub-filter to obtain best local estimation condition according to the principle of information equivalence.As the experiments proved,the FEKF SLAM not only be more robust but also has an increment in the precision so that meet the needs of navigation.(4)After the process of SLAM,we research the loop-close detect of SLAM using the method of Bag of Visual Word and framework of the Bayesian probability.Meanwhile we use the application of sparse bundle adjustment to optimize the SLAM results.