Research On Vision-lidar Fused Indoor Localization And Mapping Of Robot

Today the development of service robot is in a golden age,which is determined by technological advance and social demand.Home service robot takes the main market.As the basis of high-level functions,the autonomous movement of home service robot defines its application ways and application scenarios.But some obstacles hinder the practice of home service robot.Home service robot has the particular responsibility to ensure its accuracy of localization,safety of operation and stability of functions supplied.This paper points to localization and mapping in the such particular environment of home.Contrast to other solutions,the method presented here fuses visual images and point clouds of Li DAR to estimate the global location and construct the global consistent map.There are three main parts of this method:(1)To associate Li DAR data and camera data,an efficient extrinsic calibration method is proposed,which utilizes the motion of sensors and the common observation of trihedrons.The direction vectors of the translations and rotations of the Li DAR and the camera are extracted to calculate the relative rotation by weighted Kabsch algorithm.Based on the relative rotation,the correspondences about the edges of trihedrons in the point cloud and in the image are constructed without manual intervention.Then accurate extrinsic parameters are obtained by optimizing the distances of projections of points on trihedron edges in the point cloud to corresponding trihedron edges in the image.The algorithm has the advantages of minor calculation,easy operation,minor manual intervention and accurate results.(2)To estimate the motion of the robot and the reliable initials of the global location,key points and surfels are extracted to calculate motions.Taking the measurement errors and noises of Li DAR into account,the range image projected from the point cloud is segmented along each scan lines.Then point features and surfel features are extracted.The distance costs and implicit moving least square costs corresponding to the current point cloud handled by the distorted model of uniform motion and the previous point cloud are used to optimize the estimation of the motion.The algorithm has the advantages of features stability,motion estimation stability and minor calculation.(3)To estimate the global location of the robot and construct the global consistent map,the uncertainty model and the probability fusion of surfels are presented.The covariance matrices of surfels,namely uncertainty,originate in the analysis on errors of the measurements of Li DAR.Under the hypothesis of Gaussian distribution,the surfels whose global locations are determined are integrated into the global map upon their position and covariances.Viewing frame-to-frame motions as initials,and attaching probabilities of reliability to key points and surfels by visual semantics,the global poses are obtained by optimizing the selected distance costs between frame features and map features.The algorithm has the advantages of better precision and description of surroundings.