The Research And Realization Of Indoor Localization Algorithm Of Robot Based On Visual And Inertial Measurement Unit

Indoor positioning is needed by tremendous applications.People need to estimate the position of the carrier itself in many cases,like indoor automatic guided vehicles,hospital electronic medical guide,virtual reality,cleaning robot,family service robots and so on.There are many alternative indoor location technology solutions: GPS-based,wireless signals or scene tags based and so on.GPS-based positioning can not be applied into indoor environments,because the GPS signal is blocked and precision is low.Wireless signals or scene tags based positioning requires prior arrangement scene,which limits its expandability and ease of use.Camera and inertial measurement unit(IMU)can be used to estimate the position without relying on the external environment,which contributes to their wide application range and high accuracy.Those advantages make the positioning based on visual and IMU a good choice for mobile robot indoor positioning.Investigation and implementation of mobile robot’s indoor position algorithm is carried out by adopting visual-inertial odometry which fuses vision and inertial information.The vision-based positioning method is firstly investigated and the localization process of visual odometry is derived.In order to obtain a true scale of the positioning results,a depth camera is selected as the visual sensor.A visual odometry using depth camera is designed based on the ORB-SLAM’s algorithm framework.The designed visual odometry detects and matches the ORB feature points firstly,then uses the PnP method to minimize the reprojection errors of matched feature points.Relative motion between the consecutive cameras images is restored by designed visual odometry.After obtaining the estimated robot positioning result,a dense point cloud map of the ambient environment is reconstructed and an interactive program is designed to make quantitative analysis of the accuracy of the map.Then,experiments on a two-wheel mobile vehicle,equipped with depth camera RealSense R200 as image acquisition system,are carried out to verify the feasibility of the designed visual odometer.The positioning results and dense point cloud map’s performance are analyzed.After that,the visual-inertial odometry is investigated based on the fusion of camera and IMU.Kinetic equation of IMU represented by pre-integration is deduced.Based on ROS,a visual-inertial positioning system is built.The residual error term of the robot motion equation is constructed by using the pre-integration expressed IMU kinetic equation.The new cost function is obtained by binding this residual error term and the reprojection error term of visual odometry together.Optimize new cost function to get estimated robot’s pose,which is how visual-inertial odometry works.Finally,the optor inertial camera is adopted as the image and inertial sensor.And the feasibility of the designed visual-inertial odometry is verified by experiments on the twowheeled unmanned vehicle,after which the results are analyzed.