Research On Indoor Location Technology Based On Visual SLAM

In recent years,visual SLAM has become a very important research topic at home and abroad.This paper summarizes the current research results of visual SLAM technology,and analyses the schemes of using various visual sensors to realize SLAM.Finally,a monocular camera with the advantages of simple structure and flexible motion is selected for the research of visual SLAM indoor positioning technology.Overall,visual SLAM is divided into three main modules: front-end visual odometer,back-end optimization and loop detection.By completing these three steps,simultaneous positioning and map construction can be achieved.The problems that need to be solved in front-end visual odometer are as follows: firstly,the inter-frame motion of the camera should be calculated according to the image captured by the camera;secondly,the approximate spatial coordinates of the landmark should be estimated.The function of back-end optimization is to optimize the system and make it perform better.The function of loop detection is to correct the errors according to the previous data after the camera has reached its position in the environment,so as to make the motion trajectory and map generated by the whole SLAM process coincide with the actual situation as much as possible.Through reading and referring to a large number of documents,this paper focuses on several key issues in these three modules.The research contents include: camera imaging model and camera calibration,feature point method to estimate camera motion,back-end optimization and loop detection.A visual SLAM system based on the improved ORB feature points algorithm is studied.Finally,the experimental verification of the research system shows that it has a certain practical value.The specific research work in this paper is as follows:Firstly,the imaging model of the camera and the calibration principle of the camera are studied.On the basis of theoretical research,the calibration experiment of Zhang Zhengyou calibration method is carried out.Secondly,for the study of feature point method to estimate camera motion,SLAM requires rapidity,and FAST corner points which meet this characteristic are selected.The filtering algorithm and feature matching algorithm are optimized,and the experimental verification is carried out.Thirdly,in the real environment,there are errors in the attitude estimation of the camera in motion.In order to reduce the errors in the process,an optimization framework is used to optimize the attitude estimation.On this basis,loop detection is added to make the cameraautomatically correct when it passes through the same position.Fourthly,the positioning and map construction are studied experimentally.The proposed system is validated by experiments.The estimated camera motion trajectory obtained by experiments is compared with the actual camera motion trajectory,and the loop detection is added to the system for comparative analysis.The results show that the positioning accuracy of the system is good and it has good practicability.