Research On Mobile Robot Positioning And Navigation Based On IMU And Binocular Visual Information

The autonomous positioning and navigation technology integrating multi-sensor is one of the hot topics in the field of mobile intelligent robots.Compared with other types of sensors,vision sensors have the advantages of lower manufacturing cost,larger amount of video information collected,and intuitive images.However,it is necessary to solve the problems of image feature point loss,accuracy and robustness caused by image blur,obstacle occlusion,pure rotation and scale uncertainty caused by rapid camera rotation;The IMU can directly measure the motion velocity data of the carrier through the accelerometer and the gyroscope,provide the necessary constraints for the carrier motion,and complement the vision sensor,which can effectively solve the problem of the motion carrier under the rapid motion and the pure rotation.Based on stereo vision and IMU fusion technology,this paper aims to improve the accuracy and robustness of indoor mobile robot positioning and navigation.The main research contents include:1.Discussed the basic theory of inertial measurement positioning.Including the accelerometer and gyroscope measurement theory,the mutual conversion between coordinate systems,the four ways of robot space attitude description: rotation transformation matrix,Euler angle,rotation vector and quaternion,and further study four ways conversion relationship.2.Research on visual SLAM positioning and navigation algorithm.Firstly,the positioning principle of binocular stereo vision camera is analyzed,and the pinhole camera model and camera distortion model are established.Using Zhang Zhengyou's checkerboard calibration method,with MATLAB's own calibration kit to calibrate the binocular stereo camera,first calibrate the internal and external parameters of the two monocular cameras and the distortion parameters,and then perform stereo calibration on the binocular camera.Determine the spatial correspondence between the two monocular cameras,and then construct the PnP method through feature extraction and matching,use nonlinear optimization to solve the camera pose,and use the beam adjustment method to optimize the pose backend.3.The single-sensor positioning method based on monocular vision fusion IMU and binocular stereo vision is realized respectively.The feature matching method is used to extract and match adjacent single-frame images,and the polar geometry algorithm is used to solve the camera pose.The reasons for the failure of single sensor in the actual environment are analyzed experimentally,which lays a foundation for the research of binocular stereo vision fusion IMU sensor.4.In-depth study of inertial and binocular stereo vision fusion positioning navigation algorithm.The robustness and accuracy of the fusion data under tight coupling and loose coupling are discussed.The tight coupling is chosen as the fusion scheme.An IMU initialization method is proposed.The relative relationship of the next frame is estimated by IMU pre-integration.Velocity,gyroscope,accelerometer and gravity direction deviation;the IMU pre-integration data is combined with binocular visual information to construct a visual inertia odometer,simultaneous visual and inertial estimation error,and the carrier attitude information is determined through continuous iterative optimization.5.Using Turtlebot3 mobile robot equipped with PC and ZED binocular vision sensor to build experimental mobile platform,based on Ubuntu16.04 operating system,establish ROS robot operating environment,realize data and image visualization,and then export mobile robot pose data.Using MATLAB to simulate and analyze the data,it can be obtained from the actual error analysis.In a stable indoor environment,the fusion IMU and binocular stereo vision information can be positioned to 0.04 m,compared to single vision fusion IMU and pure binocular vision.It has higher positioning accuracy and robustness.