Research On SLAM System Based On Vision And Inertial Sensor

Real time localization and mapping(SLAM)refers to the use of various external information to determine the location of the robot in the environment,and at the same time can establish the environment map.At present,visual slam is a mature technology,but because of the complexity of the environment,a single visual sensor has been gradually difficult to meet the requirements.The low-cost inertial measurement unit(IMU)is used to assist a single visual sensor to carry out slam,and a visual inertial Sl AM(VI-SLAM)can effectively make up for the shortcomings of visual sensors.First of all,this paper studies the means of visual information measurement and processing in slam system,and studies camera model,point feature extraction algorithm,line feature extraction algorithm and parameterized model of line feature in detail,so as to use these features for feature matching between frames,construct camera residual function to get the initial estimated pose of camera,and introduce the design of word bag model and loop detection Purpose and process.Secondly,this paper studies the information measurement and processing methods of IMU,including the noise model and kinematics model of IMU,describes in detail how to use the preintegration theory of IMU to construct the residual function of IMU,and deduces the process of the variance propagation of IMU noise for back-end optimization.Then,the fusion process of vision and IMU is described in detail,and the joint initialization method of visual inertial system is deduced and given.Several threads in the system back-end optimization are introduced,including the basic principles of nonlinear optimization algorithm and loop correction.Finally,on the euroc standard data set,the algorithm designed in this paperis verified and compared with other mainstream algorithms.