Research On Indoor Positioning Technology Of Mobile Robot Based On Machine Vision And Inertial Navigation

Positioning technology is one of the key technologies to realize autonomous navigation of mobile robot.Visual SLAM is a technology for simultaneous localization and map construction,which gets a wide range of applications and researches.Aiming at the problem that visual SLAM cannot locate effectively in some indoor scenes,the relocalization is usually completed by relying on the redundancy of its own algorithm framework.However,this method leads to the decline of real-time localization,especially when the robot is moving fast.In order to avoid the impact of redundant algorithm framework on the real-time localization of visual SLAM,IMU is introduced to assist visual SLAM in this dissertation to realize the complete localization of robots under various working conditions.In this dissertation,the S1030-IR sensor with binocular camera and inertial sensor is used to study the related technologies.The main work of this dissertation is as follows:(1)The binocular camera in S1030-IR sensor is calibrated.The random error of IMU sensor is calibrated.Finally,the joint calibration of the binocular camera and the IMU sensor is carried out,and the accuracy of the calibration results is verified.The verification indicates that the calibration results meet the experimental requirements.(2)A binocular visual positioning method is proposed.The lightweight framework is established by using two threads: the tracking thread realizes the improved binocular visual odometry algorithm;The local map thread mainly completes the back-end optimization of binocular visual odometry,and obtains the optimized camera poses of key frames and optimized coordinates of map landmarks.(3)Improve the binocular visual odometry algorithm: Aiming at the problem of dense distribution of Oriented Fast key points,the homogeneous processing based on quadtree structure is presented.In order to eliminate the mismatches caused by violent matching algorithm in stereo vision,this dissertation uses RANSAC algorithm to estimate the optimal homography to remove the mismatches.Finally,the motion estimation problem of 3D-2D in the tracking process of adjacent frames is solved by using the EPNP motion estimation algorithm,and then the robot pose of the current frame is obtained.(4)Back-end optimization of the binocular visual odometry: firstly,A strategy for selecting key frames is given,and the joint optimization problem of camera poses corresponding to key frames and coordinates corresponding to map landmarks is a nonlinear least squares problem,which is solved by graph optimization method.Then two schemes that IMU assists the binocular visual positioning method in relocalization are proposed.The spatial relationship between the camera coordinate system and the IMU coordinate system at three critical moments is analyzed,and the kinematics model of the IMU corresponding to two schemes is derived and established,which solves the localization problem of the robot when the binocular visual positioning method fails to locate the robot.(5)The binocular visual positioning method is tested by using open data set.The results show that the binocular visual positioning method has high positioning accuracy and excellent real-time performance under ideal conditions,but localization failure may occur in some scenes.Based on the IMU kinematics model established in this dissertation,the position estimation and attitude estimation of the IMU are obtained in the real environment,and the experimental results prove that the attitude estimation of the IMU is more suitable for assisting the visual positioning algorithm.Experiment is carried out in the indoor scene with variational textures,and the attitude estimation of the IMU assists the binocular visual positioning method to complete relocalization.